We introduce a class of interatomic potential models that can be automatically generated from data consisting the energies and forces experienced by atoms, as derived quantum mechanical calculations. The do not have fixed functional form hence are capable modeling complex energy landscapes. They systematically improvable with more data. apply method to bulk crystals, test it calculating properties at high temperatures. Using generate long molecular dynamics trajectories required for such...

We review some recently published methods to represent atomic neighborhood environments, and analyze their relative merits in terms of faithfulness suitability for fitting potential energy surfaces. The crucial properties that such representations (sometimes called descriptors) must have are differentiability with respect moving the atoms invariance basic symmetries physics: rotation, reflection, translation, permutation same species. demonstrate certain widely used descriptors initially...

Machine learning of the quantitative relationship between local environment descriptors and potential energy surface a system atoms has emerged as new frontier in development interatomic potentials (IAPs). Here, we present comprehensive evaluation machine IAPs (ML-IAPs) based on four descriptors—atom-centered symmetry functions (ACSF), smooth overlap atomic positions (SOAP), spectral neighbor analysis (SNAP) bispectrum components, moment tensors—using diverse data set generated using...

Evaluating the (dis)similarity of crystalline, disordered and molecular compounds is a critical step in development algorithms to navigate automatically configuration space complex materials. For instance, structural similarity metric crucial for classifying structures, searching chemical better materials, driving next generation machine-learning techniques predicting stability properties molecules In last few years several strategies have been designed compare atomic coordination...

Determining the stability of molecules and condensed phases is cornerstone atomistic modelling, underpinning our understanding chemical materials properties transformations. Here we show that a machine learning model, based on local description environments Bayesian statistical learning, provides unified framework to predict atomic-scale properties. It captures quantum mechanical effects governing complex surface reconstructions silicon, predicts different classes with accuracy,...

We introduce a Gaussian approximation potential (GAP) for atomistic simulations of liquid and amorphous elemental carbon. Based on machine-learning representation the density-functional theory (DFT) potential-energy surface, such interatomic potentials enable materials with close-to DFT accuracy but at much lower computational cost. first determine maximum that any finite-range can achieve in carbon structures; then, using novel hierarchical set two-, three-, many-body structural...

Graphene nanoribbons are the counterpart of carbon nanotubes in graphene-based nanoelectronics. We investigate electronic properties chemically modified ribbons by means density functional theory. observe that chemical modifications zigzag can break spin degeneracy. This promotes onset a semiconducting-metal transition, or half-semiconducting state, with two channels having different band gap, spin-polarized where spins valence and conduction bands oppositely polarized. Edge...

We present a swift walk‐through of our recent work that uses machine learning to fit interatomic potentials based on quantum mechanical data. describe Gaussian approximation (GAP) framework, discuss variety descriptors, how train the model total energies and derivatives, simultaneous use multiple models different complexity. also show small example using QUIP, software sandbox implementation GAP is available for noncommercial use. © 2015 Wiley Periodicals, Inc.

We present the temperature dependence of growth rate carbon nanofibers by plasma-enhanced chemical vapor deposition with Ni, Co, and Fe catalysts. extrapolate a common low activation energy 0.23--0.4 eV, much lower than for thermal deposition. The diffusion on catalyst surface stability precursor molecules, ${\mathrm{C}}_{2}{\mathrm{H}}_{2}$ or ${\mathrm{CH}}_{4}$, are investigated ab initio plane wave density functional calculations. find 0.4 eV Ni Co (111) planes, bulk diffusion. barrier...

We introduce interatomic potentials for tungsten in the bcc crystal phase and its defects within Gaussian approximation potential framework, fitted to a database of first-principles density functional theory calculations. investigate performance sequence models based on databases increasing coverage configuration space showcase our strategy choosing representative small unit cells train that predict properties observable only using thousands atoms. The most comprehensive model is then used...

Statistical learning methods show great promise in providing an accurate prediction of materials and molecular properties, while minimizing the need for computationally demanding electronic structure calculations. The accuracy transferability these models are increased significantly by encoding into procedure fundamental symmetries rotational permutational invariance scalar properties. However, tensorial properties requires that model respects appropriate geometric transformations, rather...

We present an accurate machine learning (ML) model for atomistic simulations of carbon, constructed using the Gaussian approximation potential (GAP) methodology. The potential, named GAP-20, describes properties bulk crystalline and amorphous phases, crystal surfaces, defect structures with accuracy approaching that direct ab initio simulation, but at a significantly reduced cost. combine structural databases carbon graphene, which we extend substantially by adding suitable configurations,...

Amorphous silicon (a-Si) is a widely studied noncrystalline material, and yet the subtle details of its atomistic structure are still unclear. Here, we show that accurate structural models a-Si can be obtained using machine-learning-based interatomic potential. Our best network by simulated cooling from melt at rate 1011 K/s (that is, on 10 ns time scale), contains less than 2% defects, agrees with experiments regarding excess energies, diffraction data, 29Si NMR chemical shifts. We this...

The accurate representation of multidimensional potential energy surfaces is a necessary requirement for realistic computer simulations molecular systems. continued increase in power accompanied by advances correlated electronic structure methods nowadays enables routine calculations interaction energies small systems, which can then be used as references the development analytical functions (PEFs) rigorously derived from many-body (MB) expansions. Building on accuracy MB-pol PEF, we...

The allotropes of boron continue to challenge structural elucidation and solid-state theory. Here we use machine learning combined with random structure searching (RSS) algorithms systematically construct an interatomic potential for boron. Starting from ensembles randomized atomic configurations, alternating single-point quantum-mechanical energy force computations, Gaussian approximation (GAP) fitting, GAP-driven RSS iteratively generate a representation the element's potential-energy...

We present an accurate interatomic potential for graphene, constructed using the Gaussian Approximation Potential (GAP) machine learning methodology. This GAP model obtains a faithful representation of density functional theory (DFT) energy surface, facilitating highly (approaching accuracy ab initio methods) molecular dynamics simulations. is achieved at computational cost which orders magnitude lower than that comparable calculations directly invoke electronic structure methods. evaluate...

We propose a novel active learning scheme for automatically sampling minimum number of uncorrelated configurations fitting the Gaussian Approximation Potential (GAP). Our consists an unsupervised machine (ML) coupled to Bayesian optimization technique that evaluates GAP model. apply this Hafnium dioxide (HfO2) dataset generated from melt-quench ab initio molecular dynamics (AIMD) protocol. results show scheme, with no prior knowledge is able extract configuration reaches required energy fit...

Abstract In the past two and a half decades machine learning potentials have evolved from special purpose solution to broadly applicable tool for large-scale atomistic simulations. By combining efficiency of empirical force fields with an accuracy close first-principles calculations they now enable computer simulations wide range molecules materials. this perspective, we summarize present status these new types models extended systems, which are increasingly used materials modelling. There...

Many-body descriptors are widely used to represent atomic environments in the construction of machine-learned interatomic potentials and more broadly for fitting, classification, embedding tasks on structures. There is a widespread belief community that three-body correlations likely provide an overcomplete description environment atom. We produce several counterexamples this belief, with consequence any classifier, regression, or model atom-centered properties uses three- (or four)-body...