Reliable emulation of complex functionals by active learning with error control

Abstract

A statistical emulator can be used as a surrogate of complex physics-based calculations to drastically reduce the computational cost. Its successful implementation hinges on an accurate representation of the nonlinear response surface with a high-dimensional input space. Conventional "space-filling" designs, including random sampling and Latin hypercube sampling, become inefficient as the dimensionality of the input variables increases, and the predictive accuracy of the emulator can degrade substantially for a test input distant from the training input set. To address this fundamental challenge, we develop a reliable emulator for predicting complex functionals by active learning with error control (ALEC). The algorithm is applicable to infinite-dimensional mapping with high-fidelity predictions and a controlled predictive error. The computational efficiency has been demonstrated by emulating the classical density functional theory (cDFT) calculations, a statistical-mechanical method widely used in modeling the equilibrium properties of complex molecular systems. We show that ALEC is much more accurate than conventional emulators based on the Gaussian processes with "space-filling" designs and alternative active learning methods. Besides, it is computationally more efficient than direct cDFT calculations. ALEC can be a reliable building block for emulating expensive functionals owing to its minimal computational cost, controllable predictive error, and fully automatic features.

arXiv comment15 pages, 10 figures

References in corpus (21)

• Gaussian Approximation Potentials: the accuracy of quantum mechanics, without the electrons
• Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning
• SchNet - a deep learning architecture for molecules and materials
• Machine Learning Force Fields
• Moment Tensor Potentials: a class of systematically improvable interatomic potentials
• Machine Learning of Accurate Energy-Conserving Molecular Force Fields
• By-passing the Kohn-Sham equations with machine learning
• Active learning of linearly parametrized interatomic potentials
• Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields
• Finding Density Functionals with Machine Learning
• Active subspace methods in theory and practice: applications to kriging surfaces
• Perspective on integrating machine learning into computational chemistry and materials science
• Adaptive Bayesian estimation using a Gaussian random field with inverse Gamma bandwidth
• Kohn-Sham equations as regularizer: building prior knowledge into machine-learned physics
• A classical density functional from machine learning and a convolutional neural network
• Evolving symbolic density functionals
• Machine-learning free-energy functionals using density profiles from simulations
• Analytical classical density functionals from an equation learning network
• Physics-constrained Bayesian inference of state functions in classical density-functional theory
• Efficient force field and energy emulation through partition of permutationally equivalent atoms
• Monodisperse hard rods in external potentials

Cited by in corpus (9)

• Perspective: How to overcome dynamical density functional theory
• Why neural functionals suit statistical mechanics
• Hyper-density functional theory of soft matter
• Probabilistic forecast of nonlinear dynamical systems with uncertainty quantification
• Metadensity functional theory for classical fluids: Extracting the pair potential
• Learning the bulk and interfacial physics of liquid-liquid phase separation with neural density functionals
• Ab initio uncertainty quantification in scattering analysis of microscopy
• Metadensity functional learning for classical fluids: Regularizing with pair correlations
• Determining the chemical potential via universal density functional learning