About OMat24 dataset

[Asif-Iqbal-Bhatti]

Hello,

I saw that you have provided the OMat24 and sAlex datasets.

• I would like to know if the data includes all ionic trajectories or just the final geometry from VASP?
• How can one query the materials of interest just by querying the data with Li element in it.
• Is the data compatible with MP input vasp settings?

Also for training what is the best strategy to train on all ionic steps of all materials or just the final relax structure of each materials. How have you done it.

Please let me know how training is dont. I am learning this things.

Regards,
Asif

[zulissimeta]

Hi @Asif-Iqbal-Bhatti - please see the OMat24 paper for details of the dataset. You can also find details here https://fair-chem.github.io/inorganic_materials/datasets/omat24.html . The dataset is available in aselmdb format; you can see how to query results using https://wiki.fysik.dtu.dk/ase/ase/db/db.html

The most competitive approach on benchmarks like matbench-discovery right now is to train a machine learning potential on many single points, then use that potential to do a relaxation and make a final prediction. The OMat24 paper has a detailed discussion about this approach.