Size and temperature transferability of direct and local deep neural networks for atomic forces

@article{Kuritz2018SizeAT,
  title={Size and temperature transferability of direct and local deep neural networks for atomic forces},
  author={Natalia Kuritz and Goren Gordon and Amir Natan},
  journal={Physical Review B},
  year={2018}
}
A direct and local deep learning (DL) model for atomic forces is presented. We demonstrate the model performance in bulk aluminum, sodium, and silicon; and show that its errors are comparable to those found in state-of-the-art machine learning and DL models. We then analyze the model's performance as a function of the number of neighbors included and show that one can ascertain physical attributes of the system from the analysis of the deep learning model's behavior. Finally, we test the size… 

Figures from this paper

Neural network potential from bispectrum components: A case study on crystalline silicon.

TLDR
The results show that neural network potential fitting with bispectrum coefficients as descriptors is a feasible method for obtaining accurate and transferable MLFFs.

Comparison of different machine learning models for the prediction of forces in copper and silicon dioxide.

  • Wenwen LiY. Ando
  • Materials Science
    Physical chemistry chemical physics : PCCP
  • 2018
TLDR
It is found that using angular structural fingerprints and a mixture model method significantly improves the accuracy of ML force fields, and an effective structural fingerprint auto-selection method based on the least absolute shrinkage and selection operator and the genetic algorithm is discussed.

Structural analysis based on unsupervised learning: Search for a characteristic low-dimensional space by local structures in atomistic simulations

TLDR
The proposed method, which uses the two-step locality preserving projections (TS-LPP), can find a low-dimensional space wherein the distributions of datapoints for each atom or groups of atoms can be properly captured.

Machine learning for multi-fidelity scale bridging and dynamical simulations of materials

TLDR
A perspective on the challenges in multi-fidelity scale bridging is provided and the developments leading up to the use of machine learning algorithms and data-science towards addressing this grand challenge are traced.

Anharmonic thermodynamics of vacancies using a neural network potential

Lattice anharmonicity is thought to strongly affect vacancy concentrations in metals at high temperatures. It is however non-trivial to account for this effect directly using density functional

Atomistic Mechanism Underlying the Si(111)-(7×7) Surface Reconstruction Revealed by Artificial Neural-Network Potential.

TLDR
A possible step-mediated atom-pop rate-limiting process that triggers massive nonconserved atomic rearrangements, most remarkably, a critical process of collective vacancy diffusion that mediates a sequence of selective dimer, corner-hole, stacking-fault, and dimer-line pattern formation, to fulfill the 7×7 reconstruction.

Machine-learning predictions of polymer properties with Polymer Genome

TLDR
An overview of some of the critical technical aspects of Polymer Genome, including polymer data curation, representation, learning algorithms, and prediction model usage is provided, and a series of pedagogical examples are provided to demonstrate how PolymerGenome can be used to predict dozens of polymer properties, appropriate for a range of applications.

Building Nonparametric n-Body Force Fields Using Gaussian Process Regression

TLDR
This chapter presents a framework under which this problem can be tackled, based on the Bayesian construction of nonparametric force fields of a given order using Gaussian process (GP) priors, and details how properties of smoothness, invariance and interaction order of a force field can be encoded into corresponding kernel properties.

Iterative-Learning Strategy for the Development of Application-Specific Atomistic Force Fields

TLDR
A simple iterative-learning strategy for the development of machine-learning force fields targeted at specific simulations (applications) and selectively and recursively improving the force fields that are unsuitable for a given application while keeping their performance...

References

SHOWING 1-10 OF 59 REFERENCES

Machine learning for atomic forces in a crystalline solid: Transferability to various temperatures

TLDR
A machine-learning model is trained on a crystalline silicon system to directly predict the atomic forces at a wide range of temperatures using a quantum-mechanical dataset taken from canonical-ensemble simulations at a higher temperature, or an upper bound of the temperature range.

Deep Potential Molecular Dynamics: a scalable model with the accuracy of quantum mechanics

We introduce a scheme for molecular simulations, the deep potential molecular dynamics (DPMD) method, based on a many-body potential and interatomic forces generated by a carefully crafted deep

Hierarchical modeling of molecular energies using a deep neural network.

TLDR
HIP-NN achieves the state-of-the-art performance on a dataset of 131k ground state organic molecules and predicts energies with 0.26 kcal/mol mean absolute error.

Deep Potential: a general representation of a many-body potential energy surface

TLDR
Deep Potential is able to reproduce the original model, whether empirical or quantum mechanics based, within chemical accuracy, and the computational cost of this new model is not substantially larger than that of empirical force fields.

A universal strategy for the creation of machine learning-based atomistic force fields

TLDR
A general and universal strategy for using machine learning-based methods to generate highly accurate atomic force fields that may provide a pathway for performing efficient molecular dynamics simulations on nanometer-sized systems over several nanoseconds.

How to represent crystal structures for machine learning: Towards fast prediction of electronic properties

TLDR
It is found that conventional representations of the input data, such as the Coulomb matrix, are not suitable for the training of learning machines in the case of periodic solids and proposes a novel crystal structure representation for which learning and competitive prediction accuracies become possible within an unrestricted class of spd systems of arbitrary unit-cell size.

Constructing high‐dimensional neural network potentials: A tutorial review

A lot of progress has been made in recent years in the development of atomistic potentials using machine learning (ML) techniques. In contrast to most conventional potentials, which are based on

Quantum-chemical insights from deep tensor neural networks

TLDR
An efficient deep learning approach is developed that enables spatially and chemically resolved insights into quantum-mechanical observables of molecular systems, and unifies concepts from many-body Hamiltonians with purpose-designed deep tensor neural networks, which leads to size-extensive and uniformly accurate chemical space predictions.

First Principles Neural Network Potentials for Reactive Simulations of Large Molecular and Condensed Systems.

  • J. Behler
  • Computer Science
    Angewandte Chemie
  • 2017
TLDR
The methodology of an important class of ML potentials that employs artificial neural networks is considered, which can accelerate computer simulations by several orders of magnitude, while preserving quantum mechanical accuracy.

Deep learning for computational chemistry

TLDR
This review provides an introductory overview into the theory of deep neural networks and their unique properties that distinguish them from traditional machine learning algorithms used in cheminformatics, and highlights its ubiquity and broad applicability to a wide range of challenges in the field.
...