Discover ways to transform this post
POST /speech/from-post
We're starting to bring a few of the pieces together in our permanent magnet screening pipeline. In this post we'll look at how well we are able to filter out materials from a list of ~5000 ferro/ferrimagnetic materials from Materials Project.
Building property prediction models has been challenging. It's been hard to find good data for magnetic properties like anisotropy, coercivity, Curie temperature. Without access to supercomputing resources, we must find to access these properties without computing them ourselves via DFT or micro-magnetic simulations.
A quick refresher on what these properties tell us:
Remanence (Br) - residual magnetization after removal of external field
Coercivity (Hc) - resistance to demagnetization by external fields
Energy product (BHmax) - overall magnetic energy storage capacity
Curie temperature (Tc) - temperature at which magnetism is lost
Magnetocrystalline anisotropy - preferential direction of magnetization
So far, we are able to compute/predict two properties fairly well. These approaches each have their own flaws and uncertainty, but for screening, ballpark estimates are usually good enough.
Curie temperature:
In this post I'll share some of the work I've been doing on a Curie temperature prediction model. I finally found a decent dataset to work with. More on that here:
Magnetic density (often called magnetic flux density or B-field):
Magnetic density is calculated as the sum of magnetic moments per unit volume. Using CHGNet and cell volume, we can estimate this value. More specifically, the magnetic density B is related to:
B = μ₀(H + M)
Where:
B is the magnetic flux density (measured in tesla or gauss)
μ₀ is the permeability of free space
H is the external magnetic field strength
M is the magnetization of the material (the volume density of magnetic moments)
By running the ~5000 materials through both of these property predictors, we can start to filter materials that fall outside of our desired ranges.
For Curie temperature, we want to see temperatures at least above 300K room temperature, but 500+ is more desirable for practical applications.
For magnetic density, we want to see values 0.13 and above. NdBF magnets are around 0.15.
Here's a raw dataset of those results. We map Materials Project ID to their estimated properties.
A collection of 5020 magnetic materials from Materials Project, with estimated magnetic density and predicted Curie temperatures.
We will continue to build property prediction models to that we can further narrow down this list of candidates. We're still missing anisotropy/coercivity which is essential in determining a good permanent magnet candidate.
Furthermore, we should account for material's composition, filtering out materials that are high in rare-earth metals and other costly (and other negative properties like pollutive) materials.
Discover other posts like this one
To best summarize what we're looking for its worth outlining how the current state, (NdFeB) magnets, dominates and why an alternative is needed.NdFeB magnets are the strongest type of permanent magnet
In our exploration of magnetic materials, we did some embedding via Orb v2 and some dimensionality reduction.
I wanted to formalize in writing the idea that I keep coming back to for end-to-end material discovery. The hardest part of this project has been actually optimizing towards materials that have some p