Dresden 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
FM: Fachverband Funktionsmaterialien
FM 18: German-French Focus Session: (Anti)ferroic states – ferroelectrics, ferroelastics and antiferroelectrics II
FM 18.4: Vortrag
Donnerstag, 12. März 2026, 11:30–11:45, BEY/0138
MD data-driven physics-informed neural network for multi-scale modelling of ferroelectric: parameter identification and field reconstruction — •Xuejian Wang1, Frank Wendler1, Hikaru Auzuma2, and Shuji Ogata2 — 1Institute of Materials Simulation, Department of Materials Science and Engineering, Friedrich-Alexander-Universität Erlangen-Nürnberg, Fürth, Germany — 2Graduate School of Engineering, Nagoya Institute of Technology, Nagoya, Japan
A persistent challenge in multiscale ferroelectric modeling is connecting atomistic information with continuum phase-field descriptions. Here, we develop a PINN framework driven by MD polarization data. The loss function combines supervised fitting of MD-derived polarization fields with physics-based residuals of the steady-state phase-field PDEs. Minimizing the total loss enables the network to reconstruct polarization, strain, stress, and electric field distributions, while simultaneously identifying key phase-field parameters, including characteristic energy and length scales, gradient anisotropy, and Landau coefficients. Using these PINN-identified parameters in COMSOL reproduces ferroelectric domain structures and their electromechanical responses with high fidelity. This approach provides an efficient route to establishing atomistic-to-continuum links and inferring physical properties directly from polarization configurations.
Keywords: Physics-Informed Neural Network; Phase-Field Method; Molecular Dynamics Simulation; Multi-scale Model; Ferroelectrics