Ce diffusing in Na MD calculation, hard to converge

[XiangLi]

Hi all!

I am now using VASP MD to simulate the rare earth elements diffsuing in pure Na. Before doing any calculation, we need a 5ps equilibration. However I find it hard for the calculation to converge. I come up with some remedial ideas like adding the MAXMIX or set the AMIX and BMIX, it does not work very well. For 1 Ce and 127 Na atoms system, the simulation collapse within 300fs. Often return that ERROR FEXCP: supplied Exchange-correletion table is too small, maximal index : 4372 which may imply a poor model.
But the calculation succeeds in 4 Ce 127 Na atoms system.
The POSCAR is the Na bcc structure, and I replace some positions of the Na to Ce atoms.

Also I tried to simulate the pure 32 Ce fcc system in VASP MD, still it does not converge. Though without the error as above, the calculation just stops and does not move on any more at around 700fs. Here is the INCAR file I use. Also I use the pseudopotential for Ce PAW_PBE 2003 version, the latest one I think for VASP5.2.
Does anybody have some ideas? All should I add some other lines to deal with the rare earth elements like Ce? Thank you.

System =Ce-fcc Space group Fm3m
ENCUT = 500 !
LREAL = auto ! projection done in real space, fully automatic optimization of projection operators
PREC = Normal ! can choose low after tests
EDIFF = 1E-06 ! always change the default (too large drift)
ISMEAR = 1; SIGMA = 0.20 !
ALGO = Very fast ! recommended for MD (fall back ALGO = Fast A fairly robust mixutre of both algorithm is selected)
MAXMIX = 40 ! reuse mixer from one MD step to next
ISYM = 0 ! no symmetry
NELMIN =4 !
MAXMIX =180 ! about three times of the electronic steps of first iteration

IBRION = 0 ! IBRION determines how the ions are updated and moved.
NSW = 5000
NWRITE = 0
MDALGO = 2 ! Noose Hoover thermo step

TEBEG = 1000
TEEND = 1000

SMASS = 3
NBLOCK = 20
POTIM = 1

LCHARG = .TRUE.
LWAVE = .TRUE.

LASPH = .TRUE.

LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = 3
LDAUU = 5.4
LDAUJ = 0.00
LMAXMIX = 6

[Neutrino]

Hi XiangLi,

The error message

Code: Select all

ERROR FEXCP: supplied Exchange-correletion table is too small, maximal index : 4372 which may imply a poor model

indicates very high energy cuttoff supplied for a soft pseudopotential . This is documented somewhere in the forum here. The solution is to provide a harder pseudo. For Ce you can use Ce_h.

More information needed about your other simulation in order to be able to help.

Another thing, I noticed you use DFT+U for pure Ce metal. Did you try first to converge the easier DFT case?

Mostafa