Can AIMD data obtained with different EDIFF values be merged?

[qingyu_wang]

Hello everyone，
I have noticed that this webpage mentions that the merging of different ML_AB files has specific requirements—for instance, the consistency of ENCUT, k-points, and other parameters must be ensured. Given that all settings for my previous AIMD sampling are consistent except for the difference in EDIFF, can these datasets be merged? My objective is to achieve higher precision of the forces in subsequent calculations.
https://vasp.at/wiki/Best_practices_for ... rce_fields
Looking forward to your reply!

[pedro_melo]

Dear qingyu_wang,

How different are the values you used for EDIFF? Are they too different?

I would suggest to first compare the results with both values of EDIFF for a small number of structures in the data set. This will let you analyse the differences in forces, energy, and stress. Afterwards you can check whether or not these are below their expected force field errors (RMSE). If that's the case than I guess it's fine to merge your datasets.

However, note that there might still be some systematic errors that arise from not recomputing the whole set.

Kind regards,
Pedro