My Community

Good morning,

I justed started using VASP 6.4.3 and I am currently following your tutorial on Si (which is very well written tho).

I can sucessfully generate the ML_FFN from an AIMD run, but when I try to use it in the case of ionic relaxations, I get the following error:

vasp At line 2366 of file ml_ff_iohandle.F (unit = 503, file = 'ML_FF')
Fortran runtime error: End of file

So I returned to the first step, cleared it and ran once again the AIMD to get a new ML_FFN file in case the other was corrupted, didn't fix. Attached are my input files and the output containing the error.

Hoping I was clear and that you could help.

Hello!

Welcome to the VASP forum! I am sorry you are experiencing this issue when running through the tutorials. Fortunately, after trying various compilers and settings I could ultimately reproduce the error and I hope I can give you some useful advice. I believe there is indeed a bug when writing ML_FFN files in case the code was compiled without scaLAPACK. There seems to be an inconsistency between the file output and what is expected when reading in. To be honest, we rarely tested the LAPACK-only (no scaLAPACK) version of the code in the past since there are usually only disadvantages without scaLAPACK and we highly recommend to use it if available.

To confirm that my findings apply to your case, could you please send me your makefile.include and tell me which compiler you used? I suspect that the -DscaLAPACK flag is missing in the makefile.include file?

Did you avoid scaLAPACK on purpose? If not, I would just suggest to recompile VASP with scaLAPACK: basically, use one of the makefile.include templates from the arch directory and try to recompile the code (do not forget make veryclean to clean up first). If you experience problems, please post them here again! Unfortunately, you will need to recreate the ML_FFN files with the new executables before continuing with the tutorials.

Hope this helps!

All the best,
Andreas Singraber

Hey!

A short update: the bug was found and will be fixed in VASP 6.5.1.

Best,
Andreas Singraber

Hello there,

The issue came indeed from LAPACK/scaLAPACK. It has been recompiled using scaLAPACK, openMPI 2.1.6 and GCC 12.1.0, now everything works.

To answer your question about include.inc: it is implicitely managed by Spack.

Thanks a lot for your quick answer !