[Paper Review] Computational Design of the Rare-Earth Reduced Permanent Magnets

Alexander Kovacs, Johann Fischbacher+7 more|arXiv (Cornell University)|Mar 28, 2019

Magnetic Properties of Alloys24 references4 citations

TL;DR

This paper presents a computational framework combining ab initio calculations, genetic algorithms, micromagnetics, and machine learning to design rare-earth-free permanent magnets. It identifies Fe3Sn0.75Sb0.25, L10 FeNi, CoFe6Ta, and MnAl as promising candidates, predicting coercive fields of 0.49–1.0 T and energy density products up to 425 kJ/m³, demonstrating a pathway to high-performance, rare-earth-reduced magnets.

ABSTRACT

<strong>ABSTRACT</strong> Multiscale simulation is a key research tool for the quest for new permanent magnets. Starting with first principles methods, a sequence of simulation methods can be applied to calculate the maximum possible coercive field and expected energy density product of a magnet made from a novel magnetic material composition. Fe-rich magnetic phases suitable for permanent magnets can be found by adaptive genetic algorithms. The intrinsic properties computed by ab initio simulations are used as input for micromagnetic simulations of the hysteresis properties of permanent magnets with realistic structure. Using machine learning techniques, the magnet’s structure can be optimized so that the upper limits for coercivity and energy density product for a given phase can be estimated. Structure property relations of synthetic permanent magnets were computed for several candidate hard magnetic phases. The following pairs (coercive field (T), energy density product (kJ/m³)) were obtained for Fe₃Sn_0.75Sb_0.25: (0.49, 290), L1₀ FeNi: (1, 400), CoFe₆Ta: (0.87, 425), and MnAl: (0.53, 80).

Motivation & Objective

• To identify new rare-earth-free magnetic materials with high intrinsic magnetic properties suitable for permanent magnets.
• To overcome the limitations of rare-earth element dependency in high-performance magnets, which poses supply chain and cost challenges.
• To develop a predictive multiscale simulation pipeline that links atomic-scale properties to macroscopic magnet performance.
• To optimize microstructure and composition for maximum coercivity and energy product using machine learning and micromagnetic simulations.

Proposed method

• Ab initio density functional theory (DFT) calculations are used to determine the intrinsic magnetic properties of candidate Fe-rich phases.
• Adaptive genetic algorithms screen large compositional spaces to identify promising hard magnetic phases with high anisotropy and saturation magnetization.
• Micromagnetic simulations model hysteresis loops and coercive fields based on realistic microstructures derived from phase compositions.
• Machine learning models are trained on simulation data to predict optimal microstructures and estimate upper bounds for coercivity and energy product.
• A hierarchical simulation pipeline integrates results from first-principles calculations through to micromagnetic predictions, enabling efficient screening.
• The framework enables estimation of maximum achievable performance metrics for each candidate phase under idealized microstructural conditions.

Experimental results

Research questions

• RQ1 Which Fe-rich intermetallic phases exhibit sufficient magnetic anisotropy and saturation magnetization to serve as viable rare-earth-free permanent magnet materials?
• RQ2 What is the theoretical upper limit of coercive field and energy product for a given phase when microstructure is optimally designed?
• RQ3 How can machine learning accelerate the prediction of optimal microstructures for maximum magnetic performance?
• RQ4 To what extent can multiscale simulation reduce the need for experimental trial-and-error in magnet development?
• RQ5 Can genetic algorithms efficiently explore the vast compositional space of Fe-based alloys to identify high-potential candidates?

Key findings

• Fe3Sn0.75Sb0.25 is predicted to achieve a coercive field of 0.49 T and an energy density product of 290 kJ/m³.
• L10 FeNi exhibits the highest predicted performance with a coercive field of 1.0 T and an energy density product of 400 kJ/m³.
• CoFe6Ta is predicted to reach a coercive field of 0.87 T and an energy density product of 425 kJ/m³, representing a top-performing candidate.
• MnAl shows a coercive field of 0.53 T and an energy density product of 80 kJ/m³, indicating moderate performance among the candidates.
• The multiscale simulation pipeline successfully identifies performance upper bounds for each phase, enabling prioritization of experimental efforts.
• The integration of machine learning with micromagnetic simulations enables accurate estimation of optimal microstructures for maximum coercivity and energy product.