Injecting physically motivated structure into GRU-based macroscopic hysteresis models

Establish effective methods to inject physically motivated structure into single-layer GRU-based, time-resolved H-field prediction models for ferrite materials and identify which specific physical equations provide helpful inductive biases at the macroscopic scale.

Background

The study finds that a largely data-driven GRU model achieves excellent parameter efficiency, while several physics-inspired modifications (e.g., JA-based variants and PINN-style regularization) do not improve performance.

This motivates a concrete open question about how to incorporate physical knowledge in a way that demonstrably benefits macroscopic hysteresis modeling, and which equations are most suitable for this purpose.

References

As further incorporation of physically motivated ideas could not improve the performance of this model, it remains an open question how to properly inject physically motivated structure into the model and what kind of physical equations are actually helpful for the macroscopic problem at hand.

— RHINO-MAG: Recursive H-Field Inference based on Observed Magnetic Flux under Dynamic Excitation  (2603.29745 - Vater et al., 31 Mar 2026) in Section 5 Conclusion