Accurate total forces in SiO2

[bdhehr]

Dear VASP users:

I am currently attempting to calculate very accurate interatomic forces in SiO2 (quartz) for use in phonon computations. As far as I can tell, I've adjusted all of the standard INCAR switches to the settings that should minimize noise in the forces. Yet the phonon spectrum of the crystal continually exhibits a small imaginary (soft) component, and I'm not sure whether the structure is insufficiently optimized or the forces are not yet accurate enough.

The INCAR file is shown below. The electronic pressure of the relaxed structure is about -100 MPa and the residual forces are on the order of 4E-3 eV/A or less (perhaps not small enough?). I've found that augmenting kmesh or Ecut has had little effect.

Unfortunately, quartz is a relatively low-symmetry structure (four adjustable atomic coordinates and two adjustable lattice parameters) and structural optimization has proven troublesome. The relaxed structure agrees fairly well with reference data, though.

PREC = Accurate
IBRION = -1
NSW = 0
ISIF = 2
ENCUT = 650
NELM = 40
NELMIN = 2
EDIFF = 1.0e-06
EDIFFG = -0.02
VOSKOWN = 1
NBLOCK = 1
ISPIN = 1
INIWAV = 1
ISTART = 0
ICHARG = 2
LWAVE = .FALSE.
LCHARG = .TRUE.
ISMEAR = 0
SIGMA = 0.2
IALGO = 38
LREAL = .FALSE.

The simulated system contains 162 atoms and the GGA-PAW potential was used throughout.

Could the "addgrid=.TRUE." tag make a significant difference in the force accuracy here?

Thank you very much!!

[admin]

please check the setting of POTIM used for the frequencies run. It may be that the default value (0.5) is too large to ensure that you are still in the harmonic limit (if you have not set POTIM explicitely). Usually, POTIM should be set to 0.01 or 0.015 if IBRION=5