Premium
Uncovering Periodicity and Hidden Trends Responsible for Predicting the Magnetic Moment of Body Centered Cubic Crystal
Author(s) -
Takahashi Keisuke
Publication year - 2018
Publication title -
chemphyschem
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.016
H-Index - 140
eISSN - 1439-7641
pISSN - 1439-4235
DOI - 10.1002/cphc.201800141
Subject(s) - magnetic moment , moment (physics) , binary number , stability (learning theory) , inverse , magnetic susceptibility , condensed matter physics , materials science , physics , mathematics , computer science , machine learning , geometry , classical mechanics , arithmetic
Prediction of the magnetic moment of binary body centered cubic (BCC) is explored in terms of first principle calculations and data science. A dataset of 1,541 binary BCC materials constructed by first principle calculations is implemented for data mining. Descriptors for determining the magnetic moment are explored using machine learning, where classification and regression models are both implemented. Data mining reveals that two descriptors are responsible for classifying whether the materials have zero or nonzero magnetic moments and can also classify which groups of magnetic moments they belong to ( μ B < 1, 1 ≤μ B < 2, or 2 ≤μ B < 3) where the average scores produced in cross validation indicate 80 % and 91 % accuracy, respectively. Furthermore, the direct prediction of magnetic moments is performed using a regression model where eight descriptors are revealed with an average score of 74 % accuracy. The inverse problem – from a given magnetic moment to corresponding material – is successfully addressed where the stability of the predicted materials are confirmed by further first principle calculations. Thus, descriptors for the magnetic moment in BCC materials are revealed and can be seen as the base descriptor set for the magnetic moments of further complex materials.