Machine Learning Assisted Analysis of Powder Magnetic Diffuse Scattering Data

Neutron scattering has long been used as a valuable tool to study novel materials potentially characterized by exotic magnetic states. While there are a suite of neutron scattering techniques, magnetic diffuse scattering, in which neutron diffraction data is collected above the magnetic ordering temperature, has emerged as a viable technique to study the magnetic interactions in a material [1]. However, accurately determining the strength of these interactions poses significant challenges. In this talk, I will explore the information content of magnetic diffuse scattering collected on powder samples in terms of its ability to distinguish different interaction models. Inspired by recent success using artificial intelligence to assist in the analysis of neutron scattering data [2, 3], we developed a neural network architecture capable of predicting the interactions with 95% confidence for a variety of simple crystal structures with Heisenberg exchange interactions. We present a test of the network on experimental data from the diamond-lattice compound CoRh₂O₄ and were able to predict the interactions within 5% of the true value. We conclude by comparing the results obtained from machine-learning analysis with those obtained from conventional least-squares refinement.

References:

[1] T. Fennell, P. P. Deen, A. R. Wildes, et al. "Magnetic Coulomb phase in the spin ice Ho2Ti2O7". Science, 326(5951), (2009) 415-417.

[2] Z. Chen, N. Andrejevic, N. C. Drucker, et al. "Machine learning on neutron and x-ray scattering and spectroscopies" Chem. Phys. Rev. 1 (2021) 2 (3): 031301.

[3] A. M. Samarakoon, K. Barros, Y. W. Li, et al. "Machine-learning-assisted insight into spin ice Dy₂Ti₂O₇" Nat Commun 11, 892 (2020) https://doi.org/10.1038/s41467-020-14660-y

[4] J. Paddison, “Spinteract: A Program to Refine Magnetic Interactions to Diffuse Scattering Data” J. Phys.: Condens. Matter 35 (2023) 495802