My Community

Dear Mo,

I checked your input tags. You are setting ML_MODE=refit which is correct. But then you are also setting ML_ISTART = 2 which activates production mode and overwrites the ML_MODE tag. Therefore no ML_FFN file will be generated. In principle the ML_ISTART tag should not be used any more and will be replaced by ML_MODE tag.
Moreover the tag ML_FFNAME = ML_FFN does not exist in vasp. The generated ML_FF file will automatically be called ML_FFN by vasp.

All the best Jonathan

Hi again Jonathan

I've removed the tag you mentioned
Any idea why I'm still getting this rather vague error:

Machine learning selected
Setting communicators for machine learning
Initializing machine learning

===================================================================================
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= RANK 0 PID 1784541 RUNNING AT sdx20.apocrita
= KILLED BY SIGNAL: 9 (Killed)
===================================================================================

===================================================================================
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= RANK 1 PID 1784542 RUNNING AT sdx20.apocrita
= KILLED BY SIGNAL: 9 (Killed)
===================================================================================

===================================================================================
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= RANK 2 PID 1784543 RUNNING AT sdx20.apocrita
= KILLED BY SIGNAL: 9 (Killed)
===================================================================================

From this input:

#
ISTART = 1
IBRION = 0 # Conjugate gradient for stable ionic relaxation
NSW = 0 # Sufficient steps for optimization
ISIF = 3 # Relax both ions and cell shape if needed
ICHARG = 1

# Spin and Magnetism
ISPIN = 1

# MLFF Training
ML_LMLFF = .TRUE.
ML_MODE = refit
ML_MB = 2010
ML_LBASIS_DISCARD = .TRUE.
ML_MCONF_NEW = 10 # Add more configurations per iteration
ML_NMDINT = 500 # Frequency of MLFF updates
ML_CTIFOR = 0.04 # Force fitting threshold
ML_ICRITERIA = 0
ML_SCLC_CTIFOR = 0.5 # Lower to prevent overfitting
ML_WTOTEN = 1.0
ML_WTIFOR = 1.0
ML_WTSIF = 1.0
ML_RCUT1 = 10.0
ML_RCUT2 = 6.0
ML_SION1 = 0.5
ML_SION2 = 0.5
ML_MRB1 = 12
ML_MRB2 = 8
ML_W1 = 0.1
ML_LMAX2 = 3
ML_OUTBLOCK = 1
ML_OUTPUT_MODE = 1
# Molecular Dynamics Setup (if needed)
MDALGO = 3
LANGEVIN_GAMMA = 10.0 5.0 5.0 2.0 # Langevin dynamics (only if MD is used)
TEBEG = 300
TEEND = 300
#HILLS_BIN=5
#HILLS_H=0.005
#HILLS_W=0.05
#SMASS = 0

# Dispersion Correction
IVDW = 12 # DFT-D3 correction suitable for CNT interactions

# Miscellaneous
ISYM = 0 # Disable symmetry (important for doped systems)
POTIM = 0.5 # Smaller time step for stability
LREAL = .FALSE. # Accurate projection for forces/energies
LWAVE = .FALSE. # No wavefunction output unless needed
LCHARG = .FALSE. # Skip charge density file unless needed
ADDGRID = .TRUE. # Improves integration accuracy
LSCALAPACK = .FALSE. # Depending on parallelization needs
#LBLUEOUT=.TRUE.