Image file
https://magmat.herokuapp.com/benchmark
2d
I've got a small sample of experimental MAE values to compare against our calculator. While nowhere near sufficient, it should give us a bit of grounding against real world data and how trustworthy our calculator's approximations are.
Calculate Magnetic Anisotropy Energy (MAE) using DFT
We take our structures and experimental data from NovoMag. We find that their MAE calculation approach achieves an , which isn't all that great to begin with! I don't expect ours to be better given the emphasis on runtime efficiency.
On this page
https://magmat.herokuapp.com/benchmark
In the design and development of the MAE calculator here on Ouro, we focused on fast, rare-earth-free, and runnable on-demand. These constraints (and guidelines) led to a calculator, that while theoretically should work for all chemical systems, works better on certain kinds of chemical systems.
*Reran Co and MnAl with decreased kspacing (0.1, finer mesh) to match NovoMag (30x30x16), prior results was 1 MJ/m3, a major overshoot. New results are much better.
It's an interesting spread. Values track closer to experiment for some (FeNi, Fe16N2) but overshoot on CoPt.
This is likely due to the use of single-zeta basis sets. While it massively helps compute time, single-zeta basis sets are pretty minimal for capturing the orbital physics that drive MAE.
For rare-earth-free 3d magnets, SZ is more forgivable than for 4f systems since the orbitals are more localized and SOC is weaker. Still, expect 30-50% errors compared to DZP. The CoPt result supports this: the diffuse Pt 5d states with stronger SOC are more basis-sensitive.
SZ may be adequate for screening/ranking candidates in a high-throughput workflow, but quantitative predictions warrant DZP on at least the magnetic atoms.
As seen with rerunning Co, decreasing kspacing (finer mesh, increase number of kpoints), has a major effect on MAE value. This has a major impact on runtime.