Eigen RadarScience
Analysis

Magnetic reversal resistance concentrates at selected domain-wall tips in a simulation

A simulation of an idealized magnetic material locates the strongest resistance to magnetization reversal at selected domain tips and fine wall segments. The peer-reviewed work combines magnetic images with interpretable machine learning, following how domains subdivide at higher frequencies. The calculation maps a specific contribution to wall energy within the model, rather than measuring a complete device.

Science··Evening
High-angle overhead view of a computational physics worktable at the University of Tsukuba with three unlabelled simulation plates showing subdividing magnetic domains and amber-red highlights at selected domain-wall tips.

Finding resistance inside a magnet

An idealized iron-based magnet simulation places the strongest contributions to reversal resistance at the tips of growing reversed domains and selected fine wall segments. Domains are regions whose magnetization points in a common direction; walls separate neighbouring regions. The peer-reviewed paper examines these changing structures with interpretable machine learning. Japan’s University of Tsukuba also announces explainable artificial-intelligence research into the origin of magnetic fluctuations.[1], [2]

Finer domains at higher frequencies

Images from simulations on Fugaku, Japan’s supercomputer, were analysed for shape and connectivity. The team compared nominal frequencies of 100 kilohertz, 1 megahertz and 10 megahertz. One extracted coordinate follows overall magnetization, while another tracks subdivision into smaller domains. The latter correlates with the exchange-energy contribution of the walls. Mapping the features back into the images reveals that the network’s contributions are spatially uneven.[1]

A reference material in computation

Soft magnets can change their magnetization relatively easily and serve in power-conversion cores. The researchers modelled a reference structure rather than producing a commercial alloy. Their energy descriptor covers one contribution to wall energy, not every component of dissipated heat. Accelerated numerical relaxation and frequency-specific integration also restrict interpretation. The paper identifies local configurations for subsequent experimental testing with magnetic images and measured loss.[1]

References

  1. News sourceNPG Asia MaterialsMagnetic-loss analysis singles out selected domain-wall tips↩1↩2↩3
  2. News sourceUniversity of TsukubaExplainable AI identifies the origin of spatiotemporal magnetic fluctuations↩