Abstract:
Physics-Informed Neural Networks (PINNs) and Extreme Learning Machines (ELMs) are innovative numerical methods that embed physical laws directly into the training process. This enables flexible modeling of complex differential equations, such as the stray field problem in micromagnetism, without requiring extensive supervised data. The mathematical ideas of PINNs and ELMs are presented, which are employed for the full 3D
minimization of the Gibbs free energy functional. The mesh-free models eliminate the need for complex discretization schemes and are further enhanced by a hard constraint formulation using R-functions, which ensures the precise imposition of boundary conditions within the PINN framework. Additional contributions include advanced energy minimization techniques for 3D magnetization distributions, maintaining the physical
integrity of the magnetization configuration through the Cayley transform. This approach effectively models continuous magnetization distributions while minimizing total Gibbs free energy, encompassing exchange energy, anisotropy energy, and magnetostatic selfenergy.

Zoom-Link:
https://univienna.zoom.us/j/65405445489?pwd=VNlu8ESBS1IGOCifbS2YnbhiLVxkIw.1

Meeting ID: 654 0544 5489
Kenncode: 105431