In service of the main goal of discovering a rare-earth-free permanent magnet, a fast MAE predictor would be really, really helpful.
As the pipeline stands now, we have quick (high-throughput) ways of:
predicting saturation magnetization
predicting Curie temperature
predicting stability, relaxing structure
generating candidate structures (although still pretty slow)
I developed an approach with a GNN-based Hamiltonian predictor to estimate MAE from first-principles, using https://tb2j.readthedocs.io/en/latest/src/mae.html.
Unfortunately with current run times (10+ minutes per structure), it's quite inefficient to try to work directly into the pipeline where quantity is essential (search process).
I'm looking for someone to develop a new route endpoint that can take in a CIF file and output a predicted value in a more reasonable amount of time. I know it's going to be hard, but it'll be worth it.
1 awaiting review0 of 1 resolvedOpen
These older submissions are not attached to a specific item.
I've been really trying on this one! Tried a couple different approaches:
MLIP latent space as feature vector to regression model to predict MAE directly, but found no signal
HamGNN + TB2J try to calculate MAE directly, but the pretrained model I was using didn't have spin-polarization
Not a lot of luck yet, but I'm going to keep trying. Going to try ALIGNN or another GNN direct property prediction approach. More on the HamGNN approach here:
A post about trying to use HamGNN with TB2J to forecast magnetocrystalline anisotropy energy, only to find the pre-trained model lacks the needed physics. The main gap is the absence of spin-polarization in H0, making the model better suited for SOC in non-magnetic materials, not for magnetic predictions. Potential outputs still relevant to SOC include band structure corrections, topological invariants, spin textures in k-space, orbital angular momentum, spin Hall conductivity, g-factors, effective masses, and optical properties. The use case focuses on non-magnetic materials and topological insulators without magnetism. Next steps involve exploring new Hamiltonian models like DeepH-pack and MACE-H, noting they lack pre-trained models. The plan is to gather consistent data, ensure SOC and spin-polarization, and align data sources from the same DFT software. Links: https://github.com/mzjb/DeepH-pack, https://github.com/maurergroup/MACE-H.