Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 5: Machine Learning in Dynamics and Statistical Physics I

DY 5.4: Talk

Monday, March 18, 2024, 10:15–10:30, BH-N 243

Generative deep neural networks for topological defects and their microstructure reconstruction in two-dimensional spin systems — •Kyra Klos¹, Karin Everschor-Sitte², and Friederike Schmid¹ — ¹Institute of Physics, Johannes Gutenberg-University Mainz, Germany — ²Faculty of Physics and Center for Nanointegration Duisburg- Essen (CENIDE), University of Duisburg-Essen, Germany

Topological defects are stable localized perturbations of an underlying ordering field characterized by their winding number. These microscopic structures have an intrinsic multi-scale character and can be described as point like quasi-particles in the macroscopic picture. Due to long range interaction patterns and their complex implications, like the phase transition induced by topological defects, the so called Berezinskii-Kosterlitz-Thouless phase transition [1] in the two-dimensional XY Model, simulations are of high interest but difficult to realize for large system sizes. To overcome this problem we develop a generative neural network tool based on an Wasserstein Generative Adversarial Network (WGAN) [2] bridging between the microscopic and macroscopic scale. Through physics induced constraints this WGAN tool provides the opportunity to construct physical realistic representative sets of spin configuration of magnetic materials from a given defect distribution and physics input parameters.

[1] J. M. Kosterlitz, Rev. Mod. Phys. 89 (2017)

[2] M. Arjovsky et al. arXiv:1701.07875v3 (2017)

Keywords: Generative Neural Network; Topological Defects; XY Model; Machine Learning