Gaussian approximation potentials: Theory, software implementation and application examples.

Gaussian Approximation Potentials (GAPs) are a class of Machine Learned Interatomic Potentials routinely used to model materials and molecular systems on the atomic scale. The software implementation provides the means for both fitting models using ab initio data and using the resulting potentials in atomic simulations. Details of the GAP theory, algorithms and software are presented, together with detailed usage examples to help new and existing users. We review some recent developments to the GAP framework, including Message Passing Interface parallelisation of the fitting code enabling its use on thousands of central processing unit cores and compression of descriptors to eliminate the poor scaling with the number of different chemical elements.

ACEpotentials.jl: A Julia implementation of the atomic cluster expansion.

We introduce ACEpotentials.jl, a Julia-language software package that constructs interatomic potentials from quantum mechanical reference data using the Atomic Cluster Expansion [R. Drautz, Phys. Rev. B 99, 014104 (2019)]. As the latter provides a complete description of atomic environments, including invariance to overall translation and rotation as well as permutation of like atoms, the resulting potentials are systematically improvable and data efficient. Furthermore, the descriptor's expressiveness enables use of a linear model, facilitating rapid evaluation and straightforward application of Bayesian techniques for active learning. We summarize the capabilities of ACEpotentials.jl and demonstrate its strengths (simplicity, interpretability, robustness, performance) on a selection of prototypical atomistic modelling workflows.

Numerical-continuation-enhanced flexible boundary condition scheme applied to mode-I and mode-III fracture.

Motivated by the inadequacy of conducting atomistic simulations of crack propagation using static boundary conditions that do not reflect the movement of the crack tip, we extend Sinclair's flexible boundary condition algorithm [J. E. Sinclair, Philos. Mag. 31, 647 (1975)PHMAA40031-808610.1080/14786437508226544] and propose a numerical-continuation-enhanced flexible boundary scheme, enabling full solution paths for cracks to be computed with pseudo-arclength continuation, and present a method for incorporating more detailed far-field information into the model for next to no additional computational cost. The algorithms are ideally suited to study details of lattice trapping barriers to brittle fracture and can be incorporated into density functional theory and multiscale quantum and classical quantum mechanics and molecular mechanics calculations. We demonstrate our approach for mode-III fracture with a 2D toy model and employ it to conduct a 3D study of mode-I fracture of silicon using realistic interatomic potentials, highlighting the superiority of the approach over employing a corresponding static boundary condition. In particular, the inclusion of numerical continuation enables converged results to be obtained with realistic model systems containing a few thousand atoms, with very few iterations required to compute each new solution. We also introduce a method to estimate the lattice trapping range of admissible stress intensity factors K_{-}

Sensitivity and dimensionality of atomic environment representations used for machine learning interatomic potentials.

Faithfully representing chemical environments is essential for describing materials and molecules with machine learning approaches. Here, we present a systematic classification of these representations and then investigate (i) the sensitivity to perturbations and (ii) the effective dimensionality of a variety of atomic environment representations and over a range of material datasets. Representations investigated include atom centered symmetry functions, Chebyshev Polynomial Symmetry Functions (CHSF), smooth overlap of atomic positions, many-body tensor representation, and atomic cluster expansion. In area (i), we show that none of the atomic environment representations are linearly stable under tangential perturbations and that for CHSF, there are instabilities for particular choices of perturbation, which we show can be removed with a slight redefinition of the representation. In area (ii), we find that most representations can be compressed significantly without loss of precision and, further, that selecting optimal subsets of a representation method improves the accuracy of regression models built for a given dataset.

Correction: Atomistic QM/MM simulations of the strength of covalent interfaces in carbon nanotube-polymer composites.

Correction for 'Atomistic QM/MM simulations of the strength of covalent interfaces in carbon nanotube-polymer composites' by Jacek R. Golebiowski et al., Phys. Chem. Chem. Phys., 2020, 22, 12007-12014, DOI: 10.1039/d0cp01841d.

Atomistic QM/MM simulations of the strength of covalent interfaces in carbon nanotube-polymer composites.

We investigate the failure of carbon-nanotube/polymer composites by using a recently-developed hybrid quantum-mechanical/molecular-mechanical (QM/MM) approach to simulate nanotube pull-out from a cross-linked polyethene matrix. Our study focuses on the strength and failure modes of covalently-bonded nanotube-polymer interfaces based on amine, carbene and carboxyl functional groups and a [2+1] cycloaddition. We find that the choice of the functional group linking the polymer matrix to the nanotube determines the effective strength of the interface, which can be increased by up to 50% (up to the limit dictated by the strength of the polymer backbone itself) by choosing groups with higher interfacial binding energy. We rank the functional groups presented in this work based on the strength of the resulting interface and suggest broad guidelines for the rational design of nanotube functionalisation for nanotube-polymer composites.

f90wrap: an automated tool for constructing deep Python interfaces to modern Fortran codes.

f90wrap is a tool to automatically generate Python extension modules which interface to Fortran libraries that makes use of derived types. It builds on the capabilities of the popular f2py utility by generating a simpler Fortran 90 interface to the original Fortran code which is then suitable for wrapping with f2py, together with a higher-level Pythonic wrapper that makes the existance of an additional layer transparent to the final user. f90wrap has been used to wrap a number of large software packages of relevance to the condensed matter physics community, including the QUIP molecular dynamics code and the CASTEP density functional theory code.

A preconditioning scheme for minimum energy path finding methods.

Popular methods for identifying transition paths between energy minima, such as the nudged elastic band and string methods, typically do not incorporate potential energy curvature information, leading to slow relaxation to the minimum energy path for typical potential energy surfaces encountered in molecular simulation. We propose a preconditioning scheme which, combined with a new adaptive time step selection algorithm, substantially reduces the computational cost of transition path finding algorithms. We demonstrate the improved performance of our approach in a range of examples including vacancy and dislocation migration modeled with both interatomic potentials and density functional theory.

Multiscale simulations of critical interfacial failure in carbon nanotube-polymer composites.

Computational investigation of interfacial failure in composite materials is challenging because it is inherently multi-scale: the bond-breaking processes that occur at the covalently bonded interface and initiate failure involve quantum mechanical phenomena, yet the mechanisms by which external stresses are transferred through the matrix occur on length and time scales far in excess of anything that can be simulated quantum mechanically. In this work, we demonstrate and validate an adaptive quantum mechanics (QM)/molecular mechanics simulation method that can be used to address these issues and apply it to study critical failure at a covalently bonded carbon nanotube (CNT)-polymer interface. In this hybrid approach, the majority of the system is simulated with a classical forcefield, while areas of particular interest are identified on-the-fly and atomic forces in those regions are updated based on QM calculations. We demonstrate that the hybrid method results are in excellent agreement with fully QM benchmark simulations and offers qualitative insights missing from classical simulations. We use the hybrid approach to show how the chemical structure at the CNT-polymer interface determines its strength, and we propose candidate chemistries to guide further experimental work in this area.

Accelerating multiscale modelling of fluids with on-the-fly Gaussian process regression.

We present a scheme for accelerating hybrid continuum-atomistic models in multiscale fluidic systems by using Gaussian process regression as a surrogate model for computationally expensive molecular dynamics simulations. Using Gaussian process regression, we are able to accurately predict atomic-scale information purely by consideration of the macroscopic continuum-model inputs and outputs and judge on the fly whether the uncertainty of our prediction is at an acceptable level, else a new molecular simulation is performed to continually augment the database, which is never required to be complete. This provides a substantial improvement over the current generation of hybrid methods, which often require many similar atomistic simulations to be performed, discarding information after it is used once. We apply our hybrid scheme to nano-confined unsteady flow through a high-aspect-ratio converging-diverging channel, and make comparisons between the new scheme and full molecular dynamics simulations for a range of uncertainty thresholds and initial databases. For low thresholds, our hybrid solution is highly accurate-around that of thermal noise. As the uncertainty threshold is raised, the accuracy of our scheme decreases and the computational speed-up increases (relative to a full molecular simulation), enabling the compromise between accuracy and efficiency to be tuned. The speed-up of our hybrid solution ranges from an order of magnitude, with no initial database, to cases where an extensive initial database ensures no new MD simulations are required.

Machine learning unifies the modeling of materials and molecules.

Determining the stability of molecules and condensed phases is the cornerstone of atomistic modeling, underpinning our understanding of chemical and materials properties and transformations. We show that a machine-learning model, based on a local description of chemical environments and Bayesian statistical learning, provides a unified framework to predict atomic-scale properties. It captures the quantum mechanical effects governing the complex surface reconstructions of silicon, predicts the stability of different classes of molecules with chemical accuracy, and distinguishes active and inactive protein ligands with more than 99% reliability. The universality and the systematic nature of our framework provide new insight into the potential energy surface of materials and molecules.

In situ stable crack growth at the micron scale.

Grain boundaries typically dominate fracture toughness, strength and slow crack growth in ceramics. To improve these properties through mechanistically informed grain boundary engineering, precise measurement of the mechanical properties of individual boundaries is essential, although it is rarely achieved due to the complexity of the task. Here we present an approach to characterize fracture energy at the lengthscale of individual grain boundaries and demonstrate this capability with measurement of the surface energy of silicon carbide single crystals. We perform experiments using an in situ scanning electron microscopy-based double cantilever beam test, thus enabling viewing and measurement of stable crack growth directly. These experiments correlate well with our density functional theory calculations of the surface energy of the same silicon carbide plane. Subsequently, we measure the fracture energy for a bi-crystal of silicon carbide, diffusion bonded with a thin glassy layer.To improve mechanical properties in ceramics through grain boundary engineering, precise mechanical characterization of individual boundaries is vital yet difficult to achieve. Here authors perform experiments using an in situ scanning electron microscopy based double cantilever beam test, allowing to directly view and measure stable crack growth in silicon carbide.

The atomic simulation environment-a Python library for working with atoms.

The atomic simulation environment (ASE) is a software package written in the Python programming language with the aim of setting up, steering, and analyzing atomistic simulations. In ASE, tasks are fully scripted in Python. The powerful syntax of Python combined with the NumPy array library make it possible to perform very complex simulation tasks. For example, a sequence of calculations may be performed with the use of a simple 'for-loop' construction. Calculations of energy, forces, stresses and other quantities are performed through interfaces to many external electronic structure codes or force fields using a uniform interface. On top of this calculator interface, ASE provides modules for performing many standard simulation tasks such as structure optimization, molecular dynamics, handling of constraints and performing nudged elastic band calculations.

A universal preconditioner for simulating condensed phase materials.

We introduce a universal sparse preconditioner that accelerates geometry optimisation and saddle point search tasks that are common in the atomic scale simulation of materials. Our preconditioner is based on the neighbourhood structure and we demonstrate the gain in computational efficiency in a wide range of materials that include metals, insulators, and molecular solids. The simple structure of the preconditioner means that the gains can be realised in practice not only when using expensive electronic structure models but also for fast empirical potentials. Even for relatively small systems of a few hundred atoms, we observe speedups of a factor of two or more, and the gain grows with system size. An open source Python implementation within the Atomic Simulation Environment is available, offering interfaces to a wide range of atomistic codes.

Low Speed Crack Propagation via Kink Formation and Advance on the Silicon (110) Cleavage Plane.

We present density functional theory based atomistic calculations predicting that slow fracturing of silicon is possible at any chosen crack propagation speed under suitable temperature and load conditions. We also present experiments demonstrating fracture propagation on the Si(110) cleavage plane in the ~100 m/s speed range, consistent with our predictions. These results suggest that many other brittle crystals could be broken arbitrarily slowly in controlled experiments.

Molecular dynamics with on-the-fly machine learning of quantum-mechanical forces.

We present a molecular dynamics scheme which combines first-principles and machine-learning (ML) techniques in a single information-efficient approach. Forces on atoms are either predicted by Bayesian inference or, if necessary, computed by on-the-fly quantum-mechanical (QM) calculations and added to a growing ML database, whose completeness is, thus, never required. As a result, the scheme is accurate and general, while progressively fewer QM calls are needed when a new chemical process is encountered for the second and subsequent times, as demonstrated by tests on crystalline and molten silicon.

Accuracy of buffered-force QM/MM simulations of silica.

We report comparisons between energy-based quantum mechanics/molecular mechanics (QM/MM) and buffered force-based QM/MM simulations in silica. Local quantities-such as density of states, charges, forces, and geometries-calculated with both QM/MM approaches are compared to the results of full QM simulations. We find the length scale over which forces computed using a finite QM region converge to reference values obtained in full quantum-mechanical calculations is ∼10 Šrather than the ∼5 Špreviously reported for covalent materials such as silicon. Electrostatic embedding of the QM region in the surrounding classical point charges gives only a minor contribution to the force convergence. While the energy-based approach provides accurate results in geometry optimizations of point defects, we find that the removal of large force errors at the QM/MM boundary provided by the buffered force-based scheme is necessary for accurate constrained geometry optimizations where Si-O bonds are elongated and for finite-temperature molecular dynamics simulations of crack propagation. Moreover, the buffered approach allows for more flexibility, since special-purpose QM/MM coupling terms that link QM and MM atoms are not required and the region that is treated at the QM level can be adaptively redefined during the course of a dynamical simulation.

Dissociative chemisorption of O2 inducing stress corrosion cracking in silicon crystals.

Fracture experiments to evaluate the cleavage energy of the (110)[1 1 0] and (111)[1 1 2] cleavage systems in silicon at room temperature and humidity give 2.7 ± 0.3 and 2.2 ± 0.2 J/m(2), respectively, lower than any previous measurement and inconsistent with density functional theory (DFT) surface energy calculations of 3.46 and 2.88 J/m(2). However, in an inert gas environment, we measure values of 3.5 ± 0.2 and 2.9 ± 0.2 J/m(2), consistent with DFT, that suggest a previously undetected stress corrosion cracking scenario for Si crack initiation in room conditions. This is fully confirmed by hybrid quantum-mechanics-molecular-mechanics calculations.