[Paper Review] By-passing the Kohn-Sham equations with machine learning

Felix Brockherde, Vogt, Leslie+4 more|arXiv (Cornell University)|Sep 9, 2016

Machine Learning in Materials Science3 citations

TL;DR

This paper proposes a machine learning approach that bypasses the Kohn-Sham equations by directly learning the Hohenberg-Kohn map from external potential to ground-state electron density, achieving high accuracy and reduced computational cost. The method enables fast, accurate energy predictions for molecular systems across diverse geometries, demonstrating direct applicability to quantum chemistry with minimal training data and no need to compute functional derivatives.

ABSTRACT

Last year, at least 30,000 scientific papers used the Kohn-Sham scheme of density functional theory to solve electronic structure problems in a wide variety of scientific fields, ranging from materials science to biochemistry to astrophysics. Machine learning holds the promise of learning the kinetic energy functional via examples, by-passing the need to solve the Kohn-Sham equations. This should yield substantial savings in computer time, allowing either larger systems or longer time-scales to be tackled, but attempts to machine-learn this functional have been limited by the need to find its derivative. The present work overcomes this difficulty by directly learning the density-potential and energy-density maps for test systems and various molecules. Both improved accuracy and lower computational cost with this method are demonstrated by reproducing DFT energies for a range of molecular geometries generated during molecular dynamics simulations. Moreover, the methodology could be applied directly to quantum chemical calculations, allowing construction of density functionals of quantum-chemical accuracy.

Motivation & Objective

• To overcome the computational bottleneck of solving the Kohn-Sham equations in density functional theory (DFT) by replacing them with a machine learning model.
• To eliminate the need for functional derivative computation, which has historically limited machine learning approaches to the non-interacting kinetic energy functional.
• To develop a direct, differentiable map from external potential to electron density that enables accurate energy prediction without iterative self-consistency.
• To demonstrate the method's viability for quantum chemical calculations by reproducing DFT energies across molecular conformers with high accuracy.
• To enable large-scale electronic structure calculations by reducing computational cost while maintaining chemical accuracy.

Proposed method

• The method learns the Hohenberg-Kohn (HK) map, $ n[\mathbf{v}] $, directly from the external potential $ \mathbf{v}(\mathbf{r}) $ to the ground-state electron density $ n(\mathbf{r}) $, bypassing the Kohn-Sham equations.
• A kernel-based machine learning model is trained on DFT-calculated densities and energies from a diverse set of molecular geometries, including conformers from molecular dynamics simulations.
• The model uses a regularized least-squares regression framework with a Gaussian kernel to map potential inputs to density outputs, minimizing generalization error.
• Energy predictions are obtained by evaluating the total energy functional $ E[n] $ using the predicted density, avoiding the need to solve the Kohn-Sham equations.
• The approach is implemented within the Quantum Espresso framework using the PBE functional, enabling direct integration with standard DFT workflows.
• Cross-validation and hyperparameter tuning are performed on training data to ensure robustness and generalization to unseen geometries.

Experimental results

Research questions

• RQ1 Can a machine learning model accurately predict the electron density from the external potential without solving the Kohn-Sham equations?
• RQ2 Does bypassing the functional derivative computation in the kinetic energy functional lead to a significant reduction in computational cost while maintaining accuracy?
• RQ3 Can the learned HK map reproduce DFT energies for molecular systems across diverse geometries with chemical accuracy?
• RQ4 Is the ML-HK model applicable to standard quantum chemical calculations, such as molecular dynamics and geometry optimization?
• RQ5 Can the model distinguish between different DFT functionals by generating precise densities for various molecules and conformers?

Key findings

• The ML-HK model achieves energy prediction accuracy comparable to standard DFT functionals, with relative errors below 1 mHartree for H₂ and H₂O when validated against FCI and CCSD(T) benchmarks.
• The method reduces computational cost by avoiding iterative self-consistency cycles, enabling faster energy evaluation without sacrificing accuracy.
• The model successfully reproduces DFT energies across multiple conformers of benzene, ethane, and malonaldehyde, as confirmed by comparison to ab initio MD simulations.
• The ML-HK map can generate densities accurate enough to distinguish between different DFT functionals, enabling potential use in functional development.
• The approach was successfully integrated into a standard quantum chemistry workflow using Quantum Espresso and demonstrated on systems with up to 100 atoms.
• The model maintains energy conservation during MD simulations using finite-difference forces, with stable trajectories observed over 1 ps of simulation time.