[Feature Request]: Magnetic Relaxation of 2D Materials and Phonon Calculations #237
Description
Email (Optional)
Problem
Dear CHGNet Developers,
Thank you very much for developing CHGNet—it’s an impressive and powerful tool. As far as I know, it seems to be the only machine learning interatomic potential trained with magnetic moments (magmom), which is truly exciting for those of us working on magnetic materials.
I have recently been using CHGNet to optimize 2D magnetic materials. Specifically, I fixed the c-axis (to preserve vacuum spacing) and performed structure relaxations—first optimizing the lattice in-plane, then the atomic positions. Interestingly, the optimizer consistently converges to structures with force residuals (fmax) below the threshold, even for many different 2D systems. I wonder if there are any recommended settings or best practices when using CHGNet to relax 2D magnetic materials in particular.
Additionally, I’m curious whether CHGNet supports phonon calculations that self-consistently include the effect of magmom through displacements. Such functionality would be very useful and exciting。
Thanks again for your great work, and I look forward to your insights.
Chao Zhou
Proposed Solution
Add support for self-consistent phonon calculations that incorporate atomic displacements and magnetic moment responses. This would enable phonon spectrum predictions that reflect the interplay between lattice vibrations and magnetic configurations, which is particularly important for 2D magnetic materials.
Alternatives
I’ve successfully used CHGNet to relax many 2D magnetic systems by fixing the c-axis and applying in-plane lattice + atomic relaxations, achieving good convergence (fmax < threshold). However, it’s unclear whether this is the optimal approach. As for phonons, currently there seems to be no MLIP framework that self-consistently includes magmom in displaced configurations. CHGNet may be uniquely suited to fill this gap.
Code of Conduct
- I agree to follow this project's Code of Conduct