• Corpus ID: 235367854

Using differentiable programming to obtain an energy and density-optimized exchange-correlation functional

  title={Using differentiable programming to obtain an energy and density-optimized exchange-correlation functional},
  author={Sebastian Dick and Marivi Fern{\'a}ndez-Serra},
Using an end-to-end differentiable implementation of the Kohn-Sham self-consistent field equations, we obtain an accurate neural network-based exchange and correlation (XC) functional of the electronic density. The functional is optimized using information on both energy and density while exact constraints are enforced through an appropriate neural network architecture. We evaluate our model against different families of XC approximations and show that, for non-empirical functionals, there is a… 

Figures and Tables from this paper

Generalizability of density functionals learned from differentiable programming on weakly correlated spin-polarized systems

A spin-polarized version of KSR with local, semilocal, and nonlocal approximations for the exchange-correlation functional is proposed, which outperforms any existing machine learning functionals by predicting the ground-state energies of the test systems with a mean absolute error of 2.7 milli-Hartrees.

Kohn-Sham regularizer for spin density functional theory and weakly correlated systems

The authors' nonlocal functional outperforms any existing machine learning functionals by predicting the ground-state energies of the test systems with a mean absolute error of 2.7 milli-Hartrees.

D4FT: A Deep Learning Approach to Kohn-Sham Density Functional Theory

A deep learning approach to KS-DFT is proposed to directly minimize the total energy by reparameterizing the orthogonal constraint as a feed-forward computation, and it is proved that such an approach has the same expressivity as the SCF method, yet reduces the computational complexity from O(N^4) to O(3).

How Well Does Kohn-Sham Regularizer Work for Weakly Correlated Systems?

Kohn-Sham regularizer (KSR) is a differentiable machine learning approach to finding the exchange-correlation functional in Kohn-Sham density functional theory that works for strongly correlated

Automatic differentiable numerical renormalization group

Machine learning techniques have recently gained prominence in physics, yielding a host of new results and insights. One key concept is that of backpropagation, which computes the exact gradient of

Evolving symbolic density functionals

A new framework, Symbolic Functional Evolutionary Search (SyFES), is proposed that automatically constructs accurate functionals in the symbolic form, which is more explainable to humans, cheaper to evaluate, and easier to integrate to existing codes than other ML functionals.

Machine-learning-based exchange correlation functional with physical asymptotic constraints

By applying a novel ML model architecture, this study demonstrates a neural network-based exchange-correlation functional satisfying physical asymptotic constraints and improves the accuracy and generalization performance of the ML-based functional by combining the advantages of ML and analytical modeling.



Machine learning accurate exchange and correlation functionals of the electronic density

This work proposes a framework to create density functionals using supervised machine learning, termed NeuralXC, designed to lift the accuracy of baseline functionals towards that provided by more accurate methods while maintaining their efficiency.

Completing density functional theory by machine learning hidden messages from molecules

This study demonstrates that the functional can be systematically constructed using accurate density distributions and energies in reference molecules via machine learning, and will help enrich the DFT framework by utilizing the rapidly advancing machine-learning technique.

Learning from the density to correct total energy and forces in first principle simulations.

We propose a new molecular simulation framework that combines the transferability, robustness, and chemical flexibility of an ab initio method with the accuracy and efficiency of a machine learning

Climbing the density functional ladder: nonempirical meta-generalized gradient approximation designed for molecules and solids.

This work constructs a meta-GGA density functional for the exchange-correlation energy that satisfies exact constraints without empirical parameters, and describes both molecules and solids with high accuracy, as shown by extensive numerical tests.

Left-right correlation energy

We first attempt to determine a local exchange functional Ex[p] which accurately reproduces the Hartree-Fock (HF) energies of the 18 first and second row atoms. Ex[p is determined from p and |δp|,

A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu.

The revised DFT-D method is proposed as a general tool for the computation of the dispersion energy in molecules and solids of any kind with DFT and related (low-cost) electronic structure methods for large systems.

Development of the Colle-Salvetti correlation-energy formula into a functional of the electron density.

  • LeeYangParr
  • Physics
    Physical review. B, Condensed matter
  • 1988
Numerical calculations on a number of atoms, positive ions, and molecules, of both open- and closed-shell type, show that density-functional formulas for the correlation energy and correlation potential give correlation energies within a few percent.

Density functional theory is straying from the path toward the exact functional

The energy-minimizing electron densities for atomic species, as produced by 128 historical and modern DFT functionals, were found to become closer to the exact ones until the early 2000s, when this trend was reversed by unconstrained functionals sacrificing physical rigor for the flexibility of empirical fitting.

New accurate reference energies for the G2/97 test set.

This computational protocol is based on the explicitly correlated coupled-cluster method with iterative single and double excitations as well as perturbative triple excitations, using quadruple-ζ basis sets to obtain highly accurate atomization energies for the full G2/97 test set.

Toward reliable density functional methods without adjustable parameters: The PBE0 model

We present an analysis of the performances of a parameter free density functional model (PBE0) obtained combining the so called PBE generalized gradient functional with a predefined amount of exact