Unsupervised Machine Learning Techniques to explore Exotic Phases in Skyrmion Systems

F. A. GÓMEZ ALBARRACÍN

Congreso; 17th International Seminar on Condensed Matter Physics and Statistical Physics (SIMAFE); 2022

Universidad de la Frontera

(charla) Machine learning (ML) techniques have been increasingly used indifferent condensed matter areas in the last few years. Inparticular, supervised ML algorithms, such as feed forward orconvolutional neural networks, have been shown to be powerful toolsto explore and construct skyrmion phase diagrams [1][2][3]. In theseworks, significant time of preprocessing and labeling the trainingand validation data is required. On the other hand, unsupervised MLtechniques require no labeling at all. Here we apply two ofthese techniques, PCA (Principal Component Analysis) and CAE(Convolutional AutoEncoder) to detect exotic phases in skyrmionsystems. In order to do this, we first apply the chosen algorithm toan artificial dataset of configurations of ideal skyrmion lattices.Then, we apply them to configurations obtained from Monte Carlosimulations for different Hamiltonians where skyrmions phases emergeat low temperatures at intermediate magnetic fields. We show that bystudying the cross entropy and the reconstruction error it ispossible to distin-guish differentphases, such as spirals and ferromagnetic ones. Most importantly, weshow how these variables are able to suggest the emergence of exoticphenomena, not present in the more simple skyrmion models, such ashigh field bimerons or high temperature skyrmions.[1] I. A. Iakovlev,O. M. Sotnikov, and V. V. Mazurenko, Phys. Rev. B 98, 174411 (2018)[2] J.S.Salcedo-Gallo, C.C. Galindo-Gonz ́alez, E. Restrepo-Parra, J. Magn.Magn. Mat. 501, 166482 (2020).[3] F. A. GómezAlbarracín, H. D. Rosales, Phys. Rev. B 105, 214423 (2022)p { margin-bottom: 0.1in; line-height: 115%; background: transparent }