Problem With First Ionic Step

[Modey3]

Hello,

Here is the situation. I am calculating the epitaxial interfacial formation energy for the {110} BCC Fe surface mated with the {111} FCC Fe surface. I am straining the FCC slab to fit the BCC slab. First, I started out with 4-layers on each side of the epitaxial interface. The calculation structurally converged (with respect to ionic positions and lattice parameters) fine. Then I started adding more layers evenly to both sides. All the calculations converged fine expect when there were 8 layers on each side - very weird!

During the first ionic step I get the following warning:

BRMIX: very serious problems
the old and the new charge density differ
old charge density: 128.00083 new 128.00991
0.487E+01

Then after a few electronic steps I get this warning:

WARNING: Sub-Space-Matrix is not hermitian in DAV

After a 100 or so warnings like this the first ionic step converges, however the supercell pressures are very large, but the forces are reasonable. I am positive that the ionic positions are correct- after all I didn't see this problem for the smaller slabs. However, here is my POSCAR:

Epitaxial 8-layer 111 FCC Fe on 8-layer 110 BCC Fe
1
2.015 1.425 0
-2.015 1.425 0
0 0 40.947
16
Selective dynamics
cartesian
0.00000 0.00000 0 T T T
2.015 0.00000 2.015 T T T
0.00000 0.00000 4.031 T T T
2.015 0.00000 6.047 T T T
0.00000 0.00000 8.062 T T T
2.015 0.00000 10.077 T T T
0.00000 0.00000 12.092 T T T
2.015 0.00000 14.107 T T T
0.00000 0.00000 16.212 T T T
2.015 0.476 18.317 T T T
2.015 -0.476 20.422 T T T
0.00000 0.00000 22.527 T T T
2.015 0.476 24.632 T T T
2.015 -0.476 26.737 T T T
0.00000 0.00000 28.842 T T T
2.015 0.476 30.947 T T T

And here is my INCAR

SYSTEM = Fe: bcc

NSW = 50
IBRION = 2


EDIFF = 0.01
EDIFFG = -0.05

ISIF = 3

ENCUT = 260
ENAUG = 370

GGA = 91

ISPIN = 2


The nearest neighbor table looks fine. I would be thankful for any help.

Best Regards

modey3

<span class='smallblacktext'>[ Edited ]</span>

[admin]

for cells which are very elongated along one axis, it is sometimes necessary to
--reduce AMIN from the default value (which is 0.1).
also, it may be necessary to
-- reduce the mixing parameters (both for density and spin density mixing)
-- increase the k-point density: in xy, please take the converged value N obtained from your bulk convergence tests and set the k-grid to NxNx1