Parts | Days | Selection | Search | Updates | Downloads | Help
MA: Fachverband Magnetismus
MA 17: Computational Magnetism II
MA 17.6: Talk
Tuesday, March 19, 2024, 10:45–11:00, EB 202
Machine learning-based prediction of transfer integrals in undoped cuprates — •Denys Kononenko1, Ulrich K. Rößler1, Jeroen van den Brink1,2, and Oleg Janson1 — 1Institute for Theoretical Solid State Physics, IFW Dresden, Dresden, 01069, Germany — 2Institute for Theoretical Physics, TU Dresden, Dresden, 01069, Germany
Undoped cuprates represent an abundant class of magnetic insulators characterized by a complex interplay of chemistry and quantum fluctuations, resulting in diverse magnetic behaviors. Comprehending the magnetism in these materials requires understanding the underlying spin model.
Antiferromagnetic superexchange is the dominant magnetic coupling in cuprates which is estimated through electronic transfer integrals, computed using density functional theory (DFT) within the Wannier basis. However, these calculations are computationally cumbersome. We present an alternative approach based on Artificial Neural Networks (ANN) trained on high-throughput DFT calculations. The ANN predicts transfer integrals solely based on the crystal structure, offering a more efficient and less computationally demanding method. Descriptors within the ANN model capture spatial configuration and the chemical composition of the local crystalline environment.
The ANN model is a powerful tool for predicting transfer integrals and rapidly screening the relevant spin model in undoped cuprates. This development opens new avenues for designing and exploring novel materials with tailored magnetic properties.
Keywords: quantum magnetism; machine learning; high-throughput calculations; transfer integrals