Scientists at Heidelberg University have made a significant breakthrough in computational chemistry by applying machine learning to quantum chemistry research. The team has developed a new method to precisely and stably calculate molecular energies and electron densities using an orbital-free approach, which requires less computational power and enables calculations for large molecules.

The distribution of electrons in a molecule determines its chemical properties, including stability, reactivity, and biological effects. However, calculating this electron distribution and resulting energy is a central challenge in quantum chemistry. The new method, called STRUCTURES25, uses a neural network to learn the relationship between electron density and energy from precise reference calculations, capturing the chemical environment of each atom in a detailed representation.

The Heidelberg researchers trained the model with converged electron densities and variants surrounding the correct solution, generated by targeted changes in the underlying reference calculations. This approach enables the computing process to reliably find a physically meaningful solution for molecular energies and electron densities, even in cases of small deviations, and remains stable without losing its way in the calculation.

The breakthrough has significant implications for various applications, including the development of new drugs, better batteries, materials for energy conversion, and more efficient catalysts. The STRUCTURES25 method has been tested on a large and diverse collection of organic molecules, achieving a precision that can compete with established methods.

Photo by Tara Winstead on Pexels
Photos provided by Pexels