Machine Learning for Materials Design

Electronic Structure Methods and Machine Learning

Electronic Structure Methods and Machine Learning

We develop advanced electronic structure methods such as density functional theory for modeling both static and dynamic material properties. This includes creating novel exchange-correlation approximations for ground-state density functional theory, and applying time-dependent density functional theory for dynamic phenomena. We also integrate artificial intelligence techniques to enhance the accuracy and efficiency of electronic structure methods.

Dr Attila Cangi

Bartosz Brzoza
Karan Shah
Dr Lenz Fiedler
Dr Kushal Ramakrishna

Inverting the electronic structure problem with machine learning

Electronic structure theory helps us understand materials at the quantum level, supporting research in materials science and chemistry, and density functional theory is the most widely used computational method for calculating electronic structure. The accuracy of DFT relies on the exchange correlation (XC) energy and its derivative, the XC potential. Inversions can provide accurate XC potentials from known electron densities, providing insights for improved accuracy. This study explores the use of neural networks — specifically physics-informed machine learning methods — to perform these inversions. We test two approaches, physics-informed neural networks and Fourier neural operators, on one-dimensional atomic and molecular models and find that a combination of both methods has strong potential for scaling up these calculations in the future.