equilibrium geometry - convergence

[fabella]

Spin-Polarized calculation. I have relaxed the initial structure (ISPIN=1) to 0.01 eV/A. Then I read WAVECAR & CHGCAR (non-spinpol) files with ISPIN=2 and MAGMOM set correctly in the INCAR file. Now I check if spinpol equilibrium geometry is the same as non-spinpol, it is not (expected). So I relax to find equilibrium geometry for the spinpol case.

Problem! The calculation 'hangs' as if there were a segmentation fault. (?)
I originally thought I needed to change mixing parameters but after using -several- combinations this does not change the outcome. Then, I also decreased 'scaling constant for ionic steps' to POTIM=0.25. The best scenario was that one calculation did 4 ionic steps then it 'hangs' before convergence criteria satisfied.

I'm somewhat confident a VASP expert may have encountered this problem/situation and could explain this to me (I am using VASP.4.6.21).

Any suggestion what I should do? Here is what my INCAR file looks like when VASP 'hangs' (always 'hangs' just before it appears to be converging nicely).

ISTART = 1 # 0-new WAVECAR; 1-read WAVECAR
ICHARG = 1 # 1-read CHGCAR; 2-new CHGCAR
NBANDS = 248

LREAL = Auto # projector wavefunction
PREC = Normal # sets max energy cutoff for POTCAR

EDIFF = 1.0e-4 # energy convergence (eV)
NELM = 50 # max no. of scf steps

EDIFFG = -0.05 # force convergence (eV/A)
IBRION = 1 # relaxation method - quasi newton
POTIM = 0.25
IALGO = 48 # quick switch
NSW = 20 # max no. of ionic steps
NFREE = 20 # no. of ionic updates
ADDGRID = .TRUE. # add to FFT grid - accurate forces/energy

ISPIN = 2 # 2 -> spin pol., 1 -> non-spin pol. calc.
VOSKOWN = 1 # recommended spin calc. for GGA method
MAGMOM = 16*0 8*0 6*0 28*1 # set init tot. mag.mom. for each atom
# NELMDL = 0 # delay for mixing, constrain magmom 5steps
# AMIX = 0.02 # mixing for magnetic systems
# BMIX = 0.70 # significant to convergence
# AMIX_MAG = 0.20

ISMEAR = 1 # fermi smearing (partial occupancy)
SIGMA = 0.2 # smearing factor

LCHARG = .TRUE. # (default) write to CHGCAR
LWAVE = .TRUE. # (default) write to WAVECAR
NWRITE = 2 # write format to OUTCAR


W

[admin]

please specify what you mean by 'hangs'
some general remarks:
-- please check if the electronic stucture of each ionic step is really converged after 50 steps, it is a little dangerous to set a limit of electronic steps, the quality of the forces depends on the quality of electronic convegence of that step.
-- set PREC=High
-- try a different ionic relaxation mechanism (IBRION=2)
-- try ALGO=Normal
-- check if NBANDS is sufficiently high, especially for IALGO=48
-- set a higher NELMDL, (0 is heroic, even -12 may not be sufficient)

[fabella]

To say that the calculation 'hangs' it turns out that I run VASP with the ISPIN=2 tag (spin-polarized) and the calculation halts before converging with no error message or warnings. It simply stops, or just 'pauses' in that state until the running job has to be killed.

I have no idea why this -continues- to occur!

I have tried all of yours suggestions - even though I tried to be herioc a few times NELMDL led to same outcome. I turned IALGO to Normal and PREC to High as well as changed the method of relaxation (many times it doesn't even reach this point) so I am still clueless as to why this occurs?

Some notes to the environment of my calculation:
-I have a somewhat large unitcell including about 18 Angstrom of vacuum containing a large molecule and layers of metal surface. (I believe this may be a difficult system to converge to magnetic ground state!) ?

-I am running on parallel teragrid version of vasp. Does it matter?

-I really want this system to converge, this will be a prototypical model for future calculations. I need to fix this problem. What can I do to accomplish convergence?

Thank You
Best Regards,

Wesley J. Fabella
Sincere VASP user

[admin]

please check if there are major differences in the results between spin-polarized and non-polarized structures. It may be necessary to decrease the optimization level of the compilation of some subroutines