Continuing ML-FF Calculations in VASP

[santanumahapatra]

Hello,
I am trying to setup a ML-FF calculation of 147 atoms for a 10-1500K temperature ramp for first 10ps, then continuing the calculation at a constant 1500K for the next 20ps. I am using ML_ISTART=1 and ML_CTIFOR from the previous calculation to start the 2nd run. However, when I compare the OSZICAR files generated from the 2 runs (the 2nd phase is still running), I see that the 2nd phase takes significantly more time than the first.
Can you please clarify?
Do I need to copy the CONTCAR from the previous run to the POSCAR of the 2nd run, or is the ML_ISTART tag sufficient?
Thank you.

[ferenc_karsai]

First of all don't use ML_ISTART. It's a deprecated way of running MLFF calculations. Use ML_MODE (=train) instead.
It is unclear what you mean by "the second phase needs more time". I guess you mean more electronic convergence steps. That can have many reasons, but please look at your structure. It could hint that you're structure is distorted since you heat up quite strongly.
Of course if you want to continue on the trajectory from the previous run, you need to copy the CONTCAR file.

[santanumahapatra]

Thank you very much