Ab Initio Data Research Articles

Development and Validation of a ReaxFF Reactive Force Field for Modeling Silicon–Carbon Composite Anode Materials in Lithium-Ion Batteries

Silicon–carbon (Si–C) nanocomposites have recently established themselves among the most promising next-generation anode materials for lithium (Li)-ion batteries. Indeed, combined Si and graphitic (sp2-C) phases exhibit high energy densities with reasonable electrochemical responses, while covalently bonded silicon carbide (SiC) compounds offer high mechanical stiffness to withstand structural degradation. However, to form and utilize a Si–C composite structure that effectively satisfies specific battery energy, power, and lifetime requirements, it is necessary to understand the underlying atomistic mechanisms at work within its individual phases and at their interfaces. Here, we develop and validate a reactive interatomic potential (ReaxFF) for the Li–Si–C system, trained and tested against a large, diverse collection of ab initio data. In conjunction with molecular dynamics and Monte Carlo simulations, our new force field links macroscale phenomena to the nanostructures of various hybrid Si–C systems in a wide range of lithiation, temperature, and stress conditions. As an illustration, it demonstrates that the high capacity of SiC (5882 mA h g–1) is caused by amorphous lithium carbides (e.g., a-Li4.4C), which soften the overall a-Li4.4(SiC)0.5 system while swelling it volumetrically up to 668%. Furthermore, our force field accurately depicts step-wise lithiation mechanisms and volume changes in Si–sp2-C composites throughout the lithiation domain of each host subsystem. It reveals the formation of a Li-rich interphase at their grain boundary, which favors their adhesion and increases the local Li (de)insertion voltage up to 1 V. These examples demonstrate the ability of the new Li–Si–C ReaxFF potential to provide atomistic-scale insights required for designing and optimizing a wide range of Si–C-based anode candidates for upcoming Li-ion battery technologies.

New Approach To Understanding the Experimental 133Cs NMR Chemical Shift of Clay Minerals via Machine Learning and DFT-GIPAW Calculations.

Structural determination of adsorbed atoms on layered structures such as clay minerals is a complex subject. Radioactive cesium (Cs) is an important element for environmental conservation, so it is vital to understand its adsorption structure on clay. The nuclear magnetic resonance (NMR) parameters of 133Cs, which can be determined from solid-state NMR experiments, are sensitive to the local neighboring structures of adsorbed Cs. However, determining the Cs positions from NMR data alone is difficult. This paper describes an approach for identifying the expected atomic positions on clay minerals by combining machine learning (ML) with experimentally observed chemical shifts. A linear ridge regression model for ML is constructed from the smooth overlap of atomic position descriptor and gauge-including projector augmented wave (GIPAW) ab initio data. The constructed ML model predicts the GIPAW data to within a 3 ppm root-mean-squared error. At this stage, the 133Cs chemical shifts can be instantaneously calculated from the Cs positions on any clay layers using ML. The inverse analysis, which derives the atomic positions from experimentally observed chemical shifts, is developed from the ML model. The input data for the inverse analysis are the layer structure and the experimentally observed chemical shifts. The Cs positions for the targeted chemical shifts are then output. Inverse analysis is applied to montmorillonite, and the resultant Cs positions are found to be consistent with previous results (Ohkubo, T.; et al. J. Phys. Chem. A 2018, 122, 9326-9337). The Cs positions on saponite clay are also clarified from experimentally observed chemical shifts and inverse analysis.

Calculation of the Transport and Relaxation Properties of the Ar···HCl van der Waals Complex Using a New Potential Energy Surface: Comparison of the Classical and Full Quantum Mechanical Kinetic Theory Results with Molecular Dynamics Simulations.

The intermolecular potential energy surface (PES) of the Ar···HCl complex was calculated at the RCCSD(T)/aug-cc-pvQz-BF level of theory. The obtained potential was expanded in terms of Legendre polynomials and fitted to a mathematical model. The fitting results are highly correlated with the ab initio PES data with SD = 5.9 × 10-3 cm-1 and average absolute deviation (AAD) = 4.0 × 10-6 cm-1. The interaction second virial coefficients (B12) in the temperature range of 190-480 K were calculated by considering classical and first quantum corrections and compared with the available experimental data. A reasonable agreement with the experimental and calculated B12 was obtained. The PES was also used to obtain the rovibrational energy levels, and the spectroscopic rovibrational constants were obtained. It was found that the D0 values differ ∼2.25 cm-1 from the experimental values of the ground rovibrational state. Furthermore, the obtained potential was used to calculate the transport and relaxation properties using full quantum close-coupling (CC) formalism and the classical kinetic theory methods based on the Mason-Monchik approximation (MMA). It was found that the deviation between MMA and CC calculations is increased with increasing the temperature due to the higher influence of the rotational degrees of freedom on the transport properties. Also, the contribution of the inelastic (off-diagonal) transitions for diffusion coefficient is higher than the viscosity. Furthermore, the classical molecular dynamics simulations were performed using LJ(12,6) and Vashishta models, to calculate the interaction diffusion and viscosity coefficients, and compared with the results of the full quantum CC calculations. The obtained results confirm that the Vashishta model is better fitted to the ab initio potentials and is more accurate than LJ(12,6) in calculation of the diffusion coefficients.

Thermodynamic reassessment of Fe–Nb–V system

Thermodynamic reassessment of the ternary Fe–Nb–V system was carried out after a critical review of the available experimental and ab initio data. The primary solidified phases of Fe–Nb–V alloys were investigated using scanning electron microscopy and X-ray diffraction analysis. The C14 Laves phase exhibits a wide primary solidification field in the Fe-rich region, which is in contradiction with the experimental observation in the literature. Ab initio calculations were performed to obtain the enthalpies of formation of three intermetallic compounds (i.e., C14 Laves, μ, and σ), which were used to estimate the thermodynamic parameters. The end-member parameters for the three phases were improved in comparison with the previous description. The presently obtained thermodynamic description of the ternary system can reproduce the wide homogeneity range of the C14 Laves phase and the phase equilibria between the intermetallic compounds and bcc phase.

Development of accurate potentials for the physisorption of water on graphene.

From coupled-cluster singles and doubles model including connected triples corrections [CCSD(T)] calculations on the water dimer and B97D/CC on the water-circumcoronene complex at a large number of randomly generated conformations, interaction potentials for the physisorption of water on graphene are built, accomplishing almost sub-chemical accuracy. The force fields were constructed by decomposing the interaction into electrostatic and van der Waals contributions, the latter represented through improved Lennard-Jones potentials. Besides, a Chemistry at Harvard Macromolecular Mechanics (CHARMM)-like term was included in the water-water potential to improve the description of hydrogen bonds, and an induction term was added to model the polarization effects in the interaction between water and polyaromatic hydrocarbons (PAHs) or graphene. Two schemes with three and six point charges were considered for the interactions water-water and water-PAH, as Coulomb contributions are zero in the water-graphene system. The proposed fitted potentials reproduce the ab initio data used to build them in the whole range of distances and conformations and provide results for selected points very close to CCSD(T) benchmarks. When applied to the water-graphene system, the obtained results are in excellent agreement with p-CCSD(T), revised symmetry-adapted perturbation theory based on density functional theory monomer properties (DFT-SAPT), and diffusion Monte Carlo reference values. Furthermore, the stability of the various conformers water-PAH and water-graphene, as well as the different trends observed between these systems are rationalized in terms of the modifications of the electrostatic contribution.

Modeling of the interband transitions in the optical conductivity of doped two-dimensional materials in the terahertz to the infrared frequency range: the case studies of graphene and phosphorene

We use Kramers-Kronig analysis and ab initio calculations to develop a simple analytical method for including the effects of high-energy interband electron transitions in the density polarization function of doped graphene and doped phosphorene in the optical limit. The resulting formulas are suitable for applications in the terahertz to the mid-infrared range of frequencies, where the interband electron transitions are shown to give rise to static screening with a suitably chosen in-plane polarizability. In the case of phosphorene, each component of its static polarizability tensor can be computed from a sum-rule–like formula using the ab initio data for the real part of the corresponding component of the full optical interband conductivity tensor of that material.

Rapid prediction of full spin systems using uncertainty-aware machine learning.

Accurate simulation of solution NMR spectra requires knowledge of all chemical shift and scalar coupling parameters, traditionally accomplished by heuristic-based techniques or ab initio computational chemistry methods. Here we present a novel machine learning technique which combines uncertainty-aware deep learning with rapid estimates of conformational geometries to generate Full Spin System Predictions with UnCertainty (FullSSPrUCe). We improve on previous state of the art in accuracy on chemical shift values, predicting protons to within 0.209 ppm and carbons to within 1.213 ppm. Further, we are able to predict all scalar coupling values, unlike previous GNN models, achieving 3JHH accuracies between 0.838 Hz and 1.392 Hz on small experimental datasets. Our uncertainty quantification shows a strong, useful correlation with accuracy, with the most confident predictions having significantly reduced error, including our top-80% most confident proton shift predictions having an average error of only 0.140 ppm. We also properly handle stereoisomerism and intelligently augment experimental data with ab initio data through disagreement regularization to account for deficiencies in training data.

Full-dimensional automated potential energy surface development and detailed dynamics for the CH2OO + NH3 reaction.

With the help of the ROBOSURFER program package, a global full-dimensional potential energy surface (PES) for the reaction of the Criegee intermediate, CH2OO, with the NH3 molecule is developed iteratively using different ab initio methods and the monomial symmetrization fitting approach. The final permutationally-invariant analytical PES is constructed based on 23447 geometries and the corresponding ManyHF-based CCSD(T)-F12b/cc-pVTZ-F12 energies. The accuracy of the PES is confirmed by the excellent agreement of its stationary-point properties and one-dimensional potential energy curves compared with the corresponding ab initio data. The reaction probabilities and integral cross sections are calculated for the ground-state and several vibrationally excited-state reactions by quasi-classical trajectory simulations. Remarkable is that the maximum impact parameter b where reactivity vanishes is almost independent of collision energy ranging from 1 to 40 kcal mol-1, and the reaction probability increases with increasing collision energy for this negative-barrier reaction. At the same time, a slight mode-specificity effect is observed. In addition, the deuterium effect is investigated and the sudden vector projection is discussed.

Transfer learning for chemically accurate interatomic neural network potentials.

Developing machine learning-based interatomic potentials from ab initio electronic structure methods remains a challenging task for computational chemistry and materials science. This work studies the capability of transfer learning, in particular discriminative fine-tuning, for efficiently generating chemically accurate interatomic neural network potentials on organic molecules from the MD17 and ANI data sets. We show that pre-training the network parameters on data obtained from density functional calculations considerably improves the sample efficiency of models trained on more accurate ab initio data. Additionally, we show that fine-tuning with energy labels alone can suffice to obtain accurate atomic forces and run large-scale atomistic simulations, provided a well-designed fine-tuning data set. We also investigate possible limitations of transfer learning, especially regarding the design and size of the pre-training and fine-tuning data sets. Finally, we provide GM-NN potentials pre-trained and fine-tuned on the ANI-1x and ANI-1ccx data sets, which can easily be fine-tuned on and applied to organic molecules.

Force-field-enhanced neural network interactions: from local equivariant embedding to atom-in-molecule properties and long-range effects.

We introduce FENNIX (Force-Field-Enhanced Neural Network InteraXions), a hybrid approach between machine-learning and force-fields. We leverage state-of-the-art equivariant neural networks to predict local energy contributions and multiple atom-in-molecule properties that are then used as geometry-dependent parameters for physically-motivated energy terms which account for long-range electrostatics and dispersion. Using high-accuracy ab initio data (small organic molecules/dimers), we trained a first version of the model. Exhibiting accurate gas-phase energy predictions, FENNIX is transferable to the condensed phase. It is able to produce stable Molecular Dynamics simulations, including nuclear quantum effects, for water predicting accurate liquid properties. The extrapolating power of the hybrid physically-driven machine learning FENNIX approach is exemplified by computing: (i) the solvated alanine dipeptide free energy landscape; (ii) the reactive dissociation of small molecules.

New approach to the compound energy formalism (NACEF) Part II. Thermodynamic modelling of the Al–Nb system supported by first-principles calculations

In the present work, a new assessment of the Al–Nb system, using a combined first-principle and CALPHAD approach, is presented. Formation enthalpies of all ordered configurations of the intermetallic phases (σ, D022 and A15) were estimated from ab initio calculations. The liquid, fcc and bcc phases are described by a substitutional solution model. The intermetallic phases D022 and A15 are described with the New Approach to the Compound Energy Formalism (NACEF). To model the σ phase, four descriptions are applied. In first, σ phase is described with the five-sublattice (5SL) model using combined CEF model where a modification was applied which allows to respect the formation enthalpies of end-members obtained from ab initio calculations. Then, a comparison with the results obtained using the NACEF approach applied to the five-sublattice (5SL), three-sublattice (3SL) and two-sublattice (2SL) models is presented. In all cases, assessments derived from the different descriptions of σ phase show very good agreement with the available experimental and ab initio data with a limited number of used parameters. In this sense, self-consistent thermodynamic descriptions of the Al–Nb system is provided. Moreover, this work shows from the example of the σ phase that the NACEF approach, contrary to what is commonly considered, allows the compatibility between different descriptions using different numbers of sublattices. This is particularly interesting for the construction of multi-component databases.

Investigating the Heaviest Halogen: Lessons Learned from Modeling the Electronic Structure of Astatine’s Small Molecules

We present a systematic study of electron-correlation and relativistic effects in diatomic molecular species of the heaviest halogen astatine (At) within relativistic single- and multireference coupled-cluster approaches and relativistic density functional theory. We establish revised reference ab initio data for the ground states of At2, HAt, AtAu, and AtO+ using a highly accurate relativistic effective core potential model and in-house basis sets developed for accurate modeling of molecules with large spin-orbit effects. Spin-dependent relativistic effects on chemical bonding in the ground state are comparable to the binding energy or even exceed it in At2. Electron-correlation effects near the equilibrium internuclear separation are mostly dynamical and can be adequately captured using single-reference CCSD(T). However, bond elongation in At2 and, especially, AtO+ results in rapid manifestation of its multireference character. While useful for evaluating the spin-orbit effects on the ground-state bonding and properties, the two-component density functional theory lacks predictive power, especially in combination with popular empirically adjusted exchange-correlation functionals. This drawback supports the necessity to develop new functionals for reliable quantum-chemical models of heavy-element compounds with strong relativistic effects.

Optimal data generation for machine learned interatomic potentials

Machine learning interatomic potentials (MLIPs) are routinely used atomic simulations, but generating databases of atomic configurations used in fitting these models is a laborious process, requiring significant computational and human effort. A computationally efficient method is presented to generate databases of atomic configurations that contain optimal information on the small-displacement regime of the potential energy surface of bulk crystalline matter. Utilising non-diagonal supercell (Lloyd-Williams and Monserrat 2015 Phys. Rev. B 92 184301), an automatic process is suggested for ab initio data generation. MLIPs were fitted for Al, W, Mg and Si, which very closely reproduce the ab initio phonon and elastic properties. The protocol can be easily adapted to other materials and can be inserted in the workflow of any flavour of MLIP generation.

Fully coupled segregation and precipitation kinetics model with ab initio input for the Fe-Au system

For designing new and improved materials, predictive modeling of precipitation and segregation kinetics is required. So far, no modeling approach is available that couples precipitation with ab initio segregation data. We derive the mathematical basis for a model to describe segregation and precipitation kinetics with grain boundary segregation energies obtained from ab initio simulations. The model is implemented rigorously and validated with experimental data on a Fe-Au system from literature.

Piezo- and Pyroelectricity in Zirconia: A Study with Machine-Learned Force Fields

The discovery of very large piezo- and pyroelectric effects in ${\mathrm{Zr}\mathrm{O}}_{2}$ and ${\mathrm{Hf}\mathrm{O}}_{2}$-based thin films opens up opportunities to develop silicon-compatible sensor and actor devices. The effects are amplified close to the polar-orthorhombic to tetragonal phase-transition temperature. Molecular dynamics is the preferred technique to simulate such effects, though its application has to solve the dilemma between sufficient accuracy and sufficient efficiency of the interatomic force field. Here we present a deep neural-network-based interatomic force field of ${\mathrm{Zr}\mathrm{O}}_{2}$ learned from ab initio data using a systematic learning procedure in the deep potential framework. The model potential is verified to predict a variety of structural and dynamic properties with an accuracy comparable to density-functional-theory calculations. Then the deep potential model is used to reproduce the different thermal expansion and piezo- and pyroelectric phenomena in ${\mathrm{Zr}\mathrm{O}}_{2}$ with molecular dynamics calculations. At low temperature simulating the direct effect we find negative values for the piezo- and pyroelectric coefficients matching the ab initio calculations. Approaching the phase-transition temperature these values remain negative and become large. Simulating the field-induced effect above the phase-transition temperature we find positive, giant piezoelectric coefficients matching the observations. The model is able to explain the large values and the sign of the experimental observations in relation to the polar-orthorhombic to tetragonal phase transition. The model furthermore explains the recently observed giant dielectric constant in a similar system.

Determination of spectroscopic constants from rovibrational configuration interaction calculations

Rotational constants and centrifugal distortion constants of a molecule are the essence of its rotational or rovibrational spectrum (e.g., from microwave, millimeter wave, and infrared experiments). These parameters condense the spectroscopic characteristics of a molecule and, thus, are a valuable resource in terms of presenting and communicating spectroscopic observations. While spectroscopic parameters are obtained from experimental spectra by fitting an effective rovibrational Hamiltonian to transition frequencies, the ab initio calculation of these parameters is usually done within vibrational perturbation theory. In the present work, we investigate an approach related to the experimental fitting procedure, but relying solely on ab initio data obtained from variational calculations, i.e., we perform a nonlinear least squares fit of Watson's A- and S-reduced rotation-vibration Hamiltonian to rovibrational state energies (resp. transition frequencies) from rotational-vibrational configuration interaction calculations. We include up to sextic centrifugal distortion constants. By relying on an educated guess of spectroscopic parameters from vibrational configuration interaction and vibrational perturbation theory, the fitting procedure is very efficient. We observe excellent agreement with experimentally derived parameters.

Modeling the Solvation and Acidity of Carboxylic Acids Using an Ab Initio Deep Neural Network Potential.

Formic and acetic acid constitute the simplest of carboxylic acids, yet they exhibit fascinating chemistry in the condensed phase such as proton transfer and dimerization. The go-to method of choice for modeling these rare events have been accurate but expensive ab initio molecular dynamics simulations. In this study, we present a deep neural network potential trained using accurate ab initio data that can be used in tandem with enhanced-sampling methods to perform an efficient exploration of the free-energy surface of aqueous solutions of weak carboxylic acids. In particular, we show that our model captures proton dissociation and provides a good estimate of the pKa, as well as the dimerization of formic and acetic acid. This provides a suitable starting point for applications in different research areas where computational efficiency coupled with the accuracy of ab initio methods is required.

Computed spectroscopic properties of HCN, HNC, and all their D, 13C, and 15N substituted isotopologues

ABSTRACT Starting from ab initio electronic structure data, we develop parametrized analytic potential energy surfaces for the HCN and HNC isomers by variationally calculating rovibrational energy levels and adjusting the potential parameters so as to get agreement with experimentally derived transition frequencies to within about 1 cm−1. We also determine an analytic expression in terms of molecular parameters to effortlessly calculate the rovibrational energy levels. We use the obtained empirical potentials to calculate rovibrational levels for eight isotopologues of HCN and eight of HNC up to about 4000 cm−1 above the ground state. The energy levels are estimated to be accurate to within about 3 cm−1 based on comparison to experimental rovibrational transition frequencies for H12C14N, H12C14N, H13C14N, and H12C15N. For all 16 isotopologues, we calculate the zero-point energy and in nine cases we can compare with experimentally derived values. In these comparisons, the variationally obtained ZPE is within 5 cm−1 of the experimentally derived value, while the closed expression gives values within 6 cm−1 of the experimental values. For all 16 isotopologues, we also give molecular parameters from which the energy levels can easily be calculated using the closed expression. Endo- and exoergicities are given for 12 isotopic exchange reactions involving HCN/HNC and some isotopologues together with pre-exponential factors that should be useful in future modelling studies of rare isotopologues.

Orbital Mixer: Using Atomic Orbital Features for Basis-Dependent Prediction of Molecular Wavefunctions.

Leveraging ab initio data at scale has enabled the development of machine learning models capable of extremely accurate and fast molecular property prediction. A central paradigm of many previous studies focuses on generating predictions for only a fixed set of properties. Recent lines of research instead aim to explicitly learn the electronic structure via molecular wavefunctions, from which other quantum chemical properties can be directly derived. While previous methods generate predictions as a function of only the atomic configuration, in this work we present an alternate approach that directly purposes basis-dependent information to predict molecular electronic structure. Our model, Orbital Mixer, is composed entirely of multi-layer perceptrons (MLPs) using MLP-Mixer layers within a simple, intuitive, and scalable architecture that achieves competitive Hamiltonian and molecular orbital energy and coefficient prediction accuracies compared to the state-of-the-art.

Active learning of reactive Bayesian force fields applied to heterogeneous catalysis dynamics of H/Pt

Atomistic modeling of chemically reactive systems has so far relied on either expensive ab initio methods or bond-order force fields requiring arduous parametrization. Here, we describe a Bayesian active learning framework for autonomous “on-the-fly” training of fast and accurate reactive many-body force fields during molecular dynamics simulations. At each time-step, predictive uncertainties of a sparse Gaussian process are evaluated to automatically determine whether additional ab initio training data are needed. We introduce a general method for mapping trained kernel models onto equivalent polynomial models whose prediction cost is much lower and independent of the training set size. As a demonstration, we perform direct two-phase simulations of heterogeneous H2 turnover on the Pt(111) catalyst surface at chemical accuracy. The model trains itself in three days and performs at twice the speed of a ReaxFF model, while maintaining much higher fidelity to DFT and excellent agreement with experiment.