MLFF issue with ISTART =2

[ritu_tomar]

Dear Developers

I performed ml calculations with the vasp_gam version vasp.6.3.2 for an organic molecule. But I cannot perform test calculation with the ML_ISTART=2 option. The same forcefield file works if I perform further learning steps (ML_ISTART =1 works).

I encounter this error:

POSCAR found : 4 types and 59 ions
Machine learning selected
Setting communicators for machine learning
Initializing machine learning
forrtl: severe (24): end-of-file during read, unit 503, file /tmp1/tomar/HB238_molecule_crest_min__9WHr9K/ML_FF

I hope there is a solution for this. Please do let me know. Also the ML_FF file is too large even after compression to be posted here. Do let me know if I can send it some other way.

[jonathan_lahnsteiner2]

Dear ritu_tomar,

I checked your input files and tried to reproduce your error. I tried with vasp version 6.3.2 and
also with our latest vasp version 6.4.0. In both cases I was not able to reproduce your error.
I did a machine learning calculation from scratch with your input file and setting ML_ISTART=0.
Then copied the ML_FFN to ML_FF and executed a production run with ML_ISTART=2 without any problem.

I think the best option for you would be right now to switch to the latest vasp version 6.4.0 and try if
the error still occurs. It is recommended anyway to use the latest vasp version because there is an ML_MODE flag and there is also a faster execution mode which can be switched on for production runs with the ML_LFAST.

I hope this helps. And otherwise please contact us again to solve the issue

All the best Jonathan