Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MA: Fachverband Magnetismus
MA 7: Poster Magnetism I
MA 7.40: Poster
Monday, March 9, 2026, 09:30–12:30, P2
Differentiable Micromagnetics for Inverse Parameter Extraction — •Moritz Kamm1,2, Kai Litzius2, and Felix Büttner1,2 — 1Helmholtz-Zentrum Berlin für Materialien und Energie, Berlin, Germany — 2Center for Electronic Correlations and Magnetism, University of Augsburg
Extracting intrinsic micromagnetic parameters from imaging data usually involves extensive trial-and-error and often remains ambiguous. We address this challenge with a differentiable inversion method that recovers material parameters directly from a single magnetic domain image. Unlike recent neural-network approaches, the method is fully analytical and physics-driven. Because the entire pipeline is differentiable, it integrates naturally with modern optimization and gradient-based design workflows. From an experimentally accessible magnetization image, we perform a deterministic Néel-type reconstruction of the full vector field and infer material constants by enforcing fixed-point consistency under the Landau-Lifshitz-Gilbert relaxation operator. The recovery of the wall-width-determining ratio ρ=A/Ku is enabled by our autograd-compatible micromagnetic solver and consistently achieves relative errors below 1%, providing physically meaningful parameters for downstream micromagnetic modeling or the identification of candidate exotic spin textures. While demonstrated here for Néel walls, the framework is general and extensible to richer energy models, alternative lifting strategies, and other magnetic imaging modalities, establishing a foundation for differentiable inverse micromagnetics.
Keywords: Micromagnetic Simulations; Material Parameter Extraction; Differentiable Micromagnetics; Inverse Micromagnetic Modeling; Magnetic Imaging Analysis