Magnetization Reconstruction using Micromagnetics

Magnetization reconstruction has gained significant importance with the advancement of stray-field-based techniques like scanning NV magnetometry and scanning SQUID microscopy. These techniques provide unparallelly high-resolution and high sensitivity but present challenges in reconstructing the underlying magnetization. Traditional reconstruction methods, such as inverse Fourier filtering, face significant challenges due to the ill-posed nature of the backward transformation of stray-field data, often leading to artifacts and errors. While machine learning has been explored as an alternative, its “black box” nature and risk of overfitting can lead to physically meaningless reconstructions.

This study proposes an alternative methodology based on the minimization of the system’s energy by integrating a pseudo-energy term derived from stray-field measurement data into the total micromagnetic energy. Furthermore, a pseudo-effective field term is incorporated into the Landau-Lifshitz-Gilbert (LLG) equation to iteratively update magnetization directions. The feasibility of this approach is demonstrated for synthetic data with both out-of-plane and in-plane anisotropy, as well as arbitrary magnetization easy axes. We further investigate the effects of varying signal-to-noise ratios on the accuracy of the reconstruction.

Our results affirm the viability of using micromagnetics as a foundation for magnetization reconstruction, opening new approach for its application in stray-field-based techniques.