High-performance descriptor for magnetic materials: Accurate discrimination of magnetic structure

Abstract

The magnetic structure is crucial in determining the physical properties inherent in magnetic compounds. We present an adequate descriptor for magnetic structure with proper magnetic symmetry and high discrimination performance, which does not depend on artificial choices for coordinate origin, axis, and magnetic unit cell in crystal. We extend the formalism called ``smooth overlap of atomic positions'' (SOAP), providing a numerical representation of atomic configurations to that of magnetic moment configurations. We introduce the descriptor in terms of the vector spherical harmonics to describe a magnetic moment configuration and partial spectra from the expansion coefficients. We discuss that the lowest-order partial spectrum is insufficient to discriminate the magnetic structures with different magnetic anisotropy, and a higher-order partial spectrum is required in general to differentiate detailed magnetic structures on the same atomic configuration. We then introduce the fourth-order partial spectrum and evaluate the discrimination performance for different magnetic structures, mainly focusing on the difference in magnetic symmetry. The modified partial spectra that are defined not to reflect the difference of magnetic anisotropy are also useful in evaluating magnetic structures obtained from the first-principles calculations performed without spin-orbit coupling. We apply the present method to the symmetry-classified magnetic structures for the crystals of Mn${}_3$Ir and Mn${}_3$Sn, which are known to exhibit anomalous transport under the antiferromagnetic order, and examine the discrimination performance of the descriptor for different magnetic structures on the same crystal.

arXiv comment13 pages including supplementary information, 8 figures

References in corpus (10)

• Gaussian Approximation Potentials: the accuracy of quantum mechanics, without the electrons
• On representing chemical environments
• Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning
• DScribe: Library of Descriptors for Machine Learning in Materials Science
• Machine learning a general purpose interatomic potential for silicon
• Unified Representation of Molecules and Crystals for Machine Learning
• Accurate Interatomic Force Fields via Machine Learning with Covariant Kernels
• Multipole expansion for magnetic structures: A generation scheme for symmetry-adapted orthonormal basis set in crystallographic point group
• Topology analysis for anomalous Hall effect in the non-collinear antiferromagnetic states of Mn${}_{3}A$N ($A$ = Ni, Cu, Zn, Ga, Ge, Pd, In, Sn, Ir, Pt)
• A spectral-neighbour representation for vector fields: machine-learning potentials including spin

Cited by in corpus (2)

• A general formalism for machine-learning models based on multipolar-spherical harmonics
• Smooth Overlap of Spin Orientations: Machine Learning Exchange Fields for Ab-initio Spin Dynamics