Researchers have unveiled a new technique that significantly improves the accuracy of molecular modeling by refining a key component of density functional theory (DFT), a widely used simulation method in chemistry and materials science.

Computational modeling plays a crucial role in understanding materials and chemical reactions, consuming a substantial portion of supercomputer resources. The quantum many-body problem offers the highest accuracy, detailing interactions at the electron level, which governs chemical reactivity, bonding, and electrical properties. However, its computational demands limit its application to small systems.

DFT offers a more computationally efficient alternative. Instead of tracking individual electrons, DFT calculates electron densities, representing the probability of electron location in space. This allows simulations of systems with hundreds of atoms.

A significant challenge in DFT is the exchange-correlation (XC) functional, which describes electron interactions based on quantum mechanical principles. Currently, researchers rely on approximations of the XC functional tailored to specific applications.

“We know that there exists a global functional-it doesn’t matter whether the electrons are in a molecular system, a piece of metal or a semiconductor. But we do not know what its form is,” said Vikram gavini, U-M professor of mechanical engineering and corresponding author of the study in Science Advances.

The Department of Energy supported the research team’s efforts to approximate the universal XC functional, recognizing DFT’s importance for future materials and fundamental science.

the team used quantum many-body theory to study individual atoms and small molecules, effectively inverting the DFT problem. Rather of approximating the XC functional, they used machine learning to determine the XC functional that would reproduce electron behavior as calculated by quantum many-body theory.

“Many-body theories give us the right answer for the right reason, but at an unreasonable computational cost. Our team has translated many-body results into a simpler, faster form that retains most of its accuracy,” said paul Zimmerman, U-M professor of chemistry, who lead the quantum many-body calculations with chemistry Ph.D. student Jeffrey Hatch.

zimmerman’s group created a training dataset of lithium, carbon, nitrogen, oxygen, neon, dihydrogen, and lithium hydride.Adding fluorine and water did not improve the XC functional, suggesting the model had reached its limit using data from light atoms and molecules.

The resulting XC functional significantly improved DFT calculations, achieving accuracy levels beyond its complexity. DFT accuracy is frequently enough described in terms of rungs on a ladder.The team’s second-rung approach, which considers electron density gradients, achieved accuracies comparable to third-rung methods.

Implications and Future Directions

“The use of an accurate XC functional is as diverse as chemistry itself,precisely as it is indeed material agnostic.”

“The use of an accurate XC functional is as diverse as chemistry itself, precisely as it is indeed material agnostic. It’s equally relevant for researchers trying to find better battery materials to those discovering new drugs to those building quantum computers,” said Bikash Kanungo, U-M assistant research scientist in mechanical engineering and first author of the study.

Researchers can directly use the discovered XC functional or experiment with the team’s methodology. Gavini plans to extend the research to solid materials, building upon their initial work with light atoms and molecules.

The ultimate goal is to identify the universal form of the XC functional. The team aims to determine if their current functional performs well for solids, whether a new functional tailored for solids would be more effective, or if a combined functional coudl work across different material types.

Future improvements will focus on achieving higher accuracies by incorporating individual electron orbitals instead of collective electron densities.This will require significantly more computational power, given the complexity of inverting the problem at that scale.