Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 5: Machine Learning in Dynamics and Statistical Physics I
DY 5.12: Talk
Monday, March 18, 2024, 12:30–12:45, BH-N 243
Derivative learning of tensorial quantities – Predicting infrared spectra from first principles — •Bernhard Schmiedmayer1 and Georg Kresse1,2 — 1University of Vienna, Faculty of Physics and Center for Computational Materials Sciences, Vienna, Austria — 2VASP Software GmbH, Vienna, Austria
In this talk, we present a novel computational framework that integrates machine learning with first-principles calculations to achieve accurate predictions of infrared spectra. Our method demonstrates its ability to reliably generate infrared spectra for complex systems at finite temperatures. The efficiency of the method is highlighted in challenging scenarios such as the analysis of water and the organic-inorganic halide perovskite MAPbI3. Our method is in good agreement with experimental results. A unique feature of our technique is the use of derivative learning, which is essential for obtaining accurate polarization data in bulk materials and facilitates the training of a symmetry-adapted machine learning framework. Using derivative learning, we are able to predict the anti-derivative with an accuracy of about 1%.
Keywords: Symmetry-adapted machine learning; Derivative learning; Computational infrared spectra; Machine learning from first principles; Machine learning of tensorial quantities