[11a-PA8-1~6] 23.1 Joint Session N "Informatics"

[11a-PA8-4] Inverse analysis of magnetic domain structure by using persistent homology

〇Masato Kotsugi^1,3, Takumi Yamada^1,3, Chiharu Mitsumata³, Tetsuro Ueno⁶, Ippei Obayashi^3,4, Kazuto Akagi^2,3, Yasuki Hiraoka^2,3,4,5 (1.Tokyo Univ. of Sci., 2.AIMR Tohoku Univ., 3.MI2I-NIMS, 4.AIP-RIKEN, 5.Kyoto Univ., 6.QST)

Keywords:magnetic domain, persistent homology, machine learning

We demonstrated inverse analysis of magnetic domain structure to visualize the contributor of coercivity. We here applied “Persistent Homology” to the magnetic domain structure and executed the feature extraction describing macroscopic magnetic properties. Persistent Homology is a powerful methodology for quantitative evaluation of topological feature in structural data, and reduces the dimension. Moreover, it provides us the inverse analysis from magnetic property to original structural data in the combination with machine learning. We here demonstrate Persistent Homology analysis on magnetic domain structure and examine the validity and usefulness as a descriptor. As the result, we could confirm the excellent usefulness of PD as a descriptor of magnetic domains. Furthermore, we could successfully visualize the dominant contributor of coercivity in the combination with PCA.