Current status of Hamiltonian inference

Our recent paper is useful for understanding where we are. See Ref. [1] for the original paper and Ref. [2] for the source code.

Summary of what we implemented

• Atomic model for RIXS (similar to that shown previously).

• Incident-energy vs. energy-loss planes for specific geometry and incident x-ray polarization. This is the data for Fe₂O₃:[3]

• Mixture of Bayesian and a greedy optimizer PyBOBYQA for the solver

Performance

• We did 60 runs of 1000 iterations corresponding to running overnight on a personal computer.

• Obtained solutions similar to a human.

• Super-human error analysis.

---

[1]

Hamiltonian parameter inference from resonant inelastic x-ray scattering with active learning, Marton K. Lajer, Xin Dai, Kipton Barros, Matthew R. Carbone, S. Johnston, and M. P. M. Dean, Phys. Rev. B 112, 155167 (2025)

[2]

EDRIXS inverse solver

[3]

Single- and Multimagnon Dynamics in Antiferromagnetic 𝛼−Fe₂O₃ Thin Films, J. Li, Y. Gu, Y. Takahashi, K. Higashi, T. Kim, Y. Cheng, F. Yang, J. Kuneš, J. Pelliciari, A. Hariki, and V. Bisogni, Phys. Rev. X 13, 011012 (2023)