Page 1 of 1

Ionic relaxation with VASP 6.3 machine learning

Posted: Sun Nov 13, 2022 8:23 pm
by bhubnesh
Can you perform ionic relaxation calculations for larger supercells using MLFF? I Performed the calculations and my OSZICAR contains the ionic steps but not the electronic steps why? Is it written that way or I made some mistakes?

Re: Ionic relaxation with VASP 6.3 machine learning

Posted: Mon Nov 14, 2022 9:07 am
by henrique_miranda
Yes, you can perform ionic relaxation using MLFF.

Normally in ionic relaxation, you perform an SCF electronic calculation, compute the forces and update the positions such as to minimize the forces.
In the machine learning case, you start from a previously trained force field so that for each intermediate structure if you already know the forces the electronic minimization does not need to be performed.
The electronic minimization only needs to be performed if the machine-learned force field does not know the structure.

Could you please share your input files and OSZICAR and OUTCAR?

Re: Ionic relaxation with VASP 6.3 machine learning

Posted: Tue Nov 15, 2022 9:11 am
by henrique_miranda
I have seen that you posted the input files here but I did not get the chance to download them and now they are gone.
Could you please share them again?

Re: Ionic relaxation with VASP 6.3 machine learning

Posted: Tue Nov 22, 2022 9:29 pm
by bhubnesh
Hi,
Here I have attached the files.

Re: Ionic relaxation with VASP 6.3 machine learning

Posted: Wed Nov 23, 2022 8:51 am
by henrique_miranda
Have a look at the description of the ML_ISTART=2 variable:
https://www.vasp.at/wiki/index.php/ML_ISTART

In this run mode, no electronic calculations are performed and the code relies exclusively on the previously learned force field to compute the forces and total energy.