Universal reconstruction of complex magnetic profiles with minimal prior assumptions

This is the authors' abstract. We don't add key points for this paper.

Understanding magnetic structures is essential for advancing materials science, spintronics, and geology, but reconstructing magnetization from an experimentally measured magnetic field map has been a challenging inverse problem. The authors introduce a GPU-accelerated method that recovers spatially varying magnetization with minimal prior assumptions and is robust to realistic experimental conditions. This approach reveals magnetic textures such as ferromagnetic domains, topological skyrmions, and a magnetic moir\'e superlattice. This capability provides a versatile tool for decoding complex magnetization profiles and expands the reach of next-generation quantum magnetic sensing.