On this page
Automated recap of the latest activity in #permanent-magnets, created by @hermes.
Time Period: Last 26 Days
The permanent-magnets team has been actively exploring various avenues to accelerate the discovery and development of superior permanent magnets using AI-driven methodologies. The overarching goal remains clear: to create magnets that are powerful, cost-effective, easy to manufacture, and devoid of rare-earth minerals. The recent discussions and contributions from team members have centered around enhancing predictive models, leveraging novel materials strategies, and curating comprehensive datasets to support these objectives.
Several posts have delved into the challenges and advancements in building machine learning models to predict critical magnet properties, particularly Magnetocrystalline Anisotropy Energy (MAE) and Curie Temperature (T_C).
Building MAE Models:
has been actively working on developing ML models to predict MAE, recognizing its fundamental role in determining a material's ability to retain magnetization. Despite initial setbacks with existing Machine Learning Interatomic Potentials (MLIPs) like CHGNet and Orb, the exploration into specialized models such as HamGNN shows promise in capturing spin-orbit coupling effects essential for accurate MAE predictions.
Like our work on Curie temperature, the effort here is to build a machine learning model that can take a crystal structure and predict its magnetocrystalline anisotropy energy. Relevant for permanent
Universal Spin-Orbit-Coupled Hamiltonian Models:
The introduction of Uni-HamGNN by represents a significant stride towards integrating spin-orbit coupling into Hamiltonian predictions, addressing the inherent limitations of previous models in capturing complex magnetic interactions.
I came to this paper looking for a way to move beyond using a MLIP model's latent space as a feature vector to represent a material in a computational inexpensive way.
Exploration of alternative materials and doping strategies has been a focal point to circumvent the reliance on rare-earth elements.
Kinetic Hacking of Fe–Ni Magnets:
proposed an innovative approach to expedite the synthesis of tetrataenite, a high-anisotropy Fe–Ni alloy, through Hydride-Assisted Vacancy Ordering (HAVO). This method aims to achieve ordered structures in minutes rather than millennia, potentially revolutionizing the scalability of Fe–Ni magnets.
Instead of searching for new chemistries and crystal structures, we can focus on a known good material which is hard to synthesize at scale. This is an alternative path forward to an idea like this on
Heavy-p “SOC-Donor” Magnets:
Another significant contribution by involves leveraging heavy p-block elements (e.g., Se, Te, Sb, Bi) to donate spin-orbit coupling to 3d transition metals. This strategy aims to enhance magnetocrystalline anisotropy without incorporating rare-earth elements, opening pathways to new classes of high-performance, rare-earth-free magnets.
Rare-earth elements earned their place in permanent magnets because the large atomic spin-orbit coupling (SOC) of the 4 f shell turns exchange energy into a hefty magnetocrystalline anisotropy (MAE).
Robust datasets are critical for training accurate predictive models. Efforts have been concentrated on aggregating and cleaning data related to magnet properties.
Novomag and Novamag Databases:
has been meticulously curating data from the Novomag and Novamag databases, which offer extensive datasets on magnetic materials, including MAE, saturation magnetization, and Curie temperatures. These resources are invaluable for building and validating ML models aimed at predicting magnet properties.
Also known as the Magnetic Materials Database. I came to this database looking for magnetocrystalline anisotropy energy data for permanent magnet design. After scraping the data from the app, which is
MAGNDATA and NEMAD:
Additional datasets like MAGNDATA and NEMAD are being explored to supplement the existing information pool, although challenges remain in extracting and standardizing MAE-related data from these sources.
Working on cleaning the data we have available and seeing what we've got for a MAE prediction model. This resource was nice and had all the raw files uploaded so that you can process them yourself and
The synergy between AI models and material synthesis techniques is crucial for translating computational predictions into tangible magnets.
Design Rules for Iron Magnets:
Insights into structuring iron-based magnets, such as maintaining short axial bonds and incorporating heavy elements to boost spin-orbit coupling, provide actionable guidelines for synthesizing high-MAE materials.
Sharing some notes as I go through this paper:
Accurate MAE Prediction:
Current ML models, particularly standard MLIPs, fall short in capturing the intricacies of spin-orbit coupling essential for MAE. Developing specialized models like Uni-HamGNN is imperative.
Dataset Integration and Standardization:
Consolidating data from multiple sources (Novomag, Novamag, MAGNDATA, NEMAD) remains a challenge due to inconsistencies in data formats and completeness, especially concerning MAE values.
Scalable Synthesis Methods:
Translating AI-driven material predictions into scalable synthesis processes, such as the proposed HAVO method for Fe–Ni magnets, requires access to specialized equipment and validation in experimental setups.
Balancing Magnet Properties:
Achieving high MAE without compromising other essential properties like saturation magnetization and Curie temperature is a delicate balance that necessitates nuanced material design strategies.
Collaborative Integration:
There is potential for deeper collaboration between modelers and experimentalists to ensure that computational predictions are aligned with practical synthesis capabilities.
Advanced ML Models:
Exploring state-of-the-art models that inherently incorporate physical laws related to magnetism could bridge the gap between prediction accuracy and computational efficiency.
Benchmarking and Validation:
Establishing robust benchmarking protocols to validate model predictions against experimental data will enhance the reliability of AI-driven discoveries.
Exploration of New Material Spaces:
Leveraging AI to navigate less-explored material families, as suggested in the heavy-p donor strategy, could uncover novel magnets with superior properties.
The permanent-magnets team is making commendable progress in harnessing AI for the discovery of rare-earth-free permanent magnets. By addressing the challenges in predictive modeling and dataset curation, and by exploring innovative materials strategies, the team is well-positioned to achieve its goals. Continued emphasis on interdisciplinary collaboration and model refinement will be pivotal in overcoming current hurdles and realizing the vision of scalable, high-performance permanent magnets.