My Community

Dear all I am trying to optimize a slab of Anatase001. It is constituted by 4 layers of anatase. I initially "freeze" the whole structure. Then I "unfreeze" layer by layer. I want to simulate an asymmetric slab with the ions in the first (bottom) layer kept in bulk like position.

I let optimize the upper layer, keeping "frozen" the remaining 3. I get convergence after 33 cycles. Then I "unfreeze" also the second layer. But now after 500 cycles I do not reach convergency. I restart. Others 500 cycles. I restart a third time and immediately I get the following:
....
WARNING: Sub-Space-Matrix is not hermitian in DAV 21
-1189581.37010824
x is not hermitian in DAV 21
-1189581.37010824
DAV: 92 -0.645591984032E+06 -0.25580E+06 -0.30242E+05 3832 0.365E+03BRMIX: very serious problems
the old and the new charge density differ
old charge density: 132.91428 new 140.50681
0.330E+02
WARNING: Sub-Space-Matrix is not hermitian in DAV 8
6.701766384101552E-003
WARNING: Sub-Space-Matrix is not hermitian in DAV 8
6.701766384101552E-003
WARNING: Sub-Space-Matrix is not hermitian in DAV 8
6.701766384101552E-003
Error EDDDAV: Call to ZHEGV failed. Returncode = 74 5 6

.....

I almost know the meaning of this message even if I do not understand its reason since I have not changed any parameters in INCAR with respect to the previous two optimizations
Nevertheless, I wonder if there is someone who can give me some hints for achieving the convergence. I attach INCAR and POSCAR (KPOINTS is 5x5x1 Gamma centered).

Best,
G


INCAR

SYSTEM = Si_Vac_Td

# xc
GGA = PE
VOSKOWN = 1
ISTART = 1

#electronic relaxation
#ALGO = Fast
LREAL = Auto
PREC= Medium

ISMEAR = 0
SIGMA = 0.05

#electronic RELAXATION
NELMIN = 4
NELM = 200

ISIF = 2

IBRION = 1

IMIX = 4 ! Pulay
WC = 1000.
AMIX = 0.3 ! Mix coefficient with new function
MAXMIX = -25 ! States stored in Pulay for electronic step

EDIFF = 1.E-5
EDIFFG = -0.02

NSW = 500




POSCAR

Ti O
1.00000000000000
3.785 0.000 0.000
0.000 3.785 0.000
0.000 0.000 58.056
16 32
Selective dynamics
Direct
0.00 0.000 0.1722474851867162 F F F
0.00 0.500 0.2132165495383802 F F F
0.50 0.500 0.2541856138900371 F F F
0.50 0.000 0.2951546782417012 F F F
0.00 0.000 0.3361237425933581 F F F
0.00 0.500 0.3770928069450221 F F F
0.50 0.500 0.4180618712966790 F F F
0.50 0.000 0.4590309356483431 F F F
0.00 0.000 0.5002213728129576 T T T
0.00 0.50000 0.5417804080055326 T T T
0.50 0.50000 0.5834417961183826 T T T
0.500 0.0000 0.6252200095199100 T T T
0.000 0.00000 0.6672280134917706 T T T
0.000 0.50000 0.7095621547392290 T T T
0.5000 0.5000 0.7520075712738961 T T T
0.5000 0.0000 0.7932327080769714 T T T
0.0000 0.5000 0.1789457509990342 F F F
0.0000 0.0000 0.2065182492765629 F F F
0.5000 0.5000 0.2199148498001904 F F F
0.0000 0.5000 0.2474873136282199 F F F
0.50 0.00 0.2608839141518544 F F F
0.50 0.50 0.2884563779798839 F F F
0.00 0.00 0.3018529785035113 F F F
0.50 0.00 0.3294254423315408 F F F
0.00 0.50 0.3428220084056761 F F F
0.00 0.00 0.3703945066832048 F F F
0.50 0.50 0.3837911072068323 F F F
0.00 0.50 0.4113635710348618 F F F
0.50 0.00 0.4247601715584963 F F F
0.50 0.50 0.4523326353865258 F F F
0.00 0.00 0.4657292359101532 F F F
0.50 0.00 0.4933016997381827 F F F
0.00 0.50 0.5072660043548354 T T T
0.00 0.00 0.5347205387894126 T T T
0.50 0.50 0.5489205293895518 T T T
0.00 0.50 0.5763880449558654 T T T
0.50 0.00 0.5905946023173325 T T T
0.50 0.50 0.6180578994192893 T T T
0.00 0.00 0.6325916796571955 T T T
0.50 0.00 0.6597324773931698 T T T
0.00 0.50 0.6748998564400435 T T T
0.00 0.00 0.7018883718261607 T T T
0.50 0.50 0.7172747705492182 T T T
0.00 0.50 0.7443535251509735 T T T
0.50 0.00 0.7595784027756771 T T T
0.50 0.50 0.7862599499087540 T T T
0.00 0.00 0.8020845216039235 T T T
0.50 0.00 0.1655491849248989 F F F

Did you remove your old WAVECAR? It seems you write it.
Do not change the mixing parameters! The standard works for that case unless sth. is screwed.
You did not set ENCUT. Doit. Just for safety.

Hth

alex

the message concerning the charge neurtality indicates that sth
is severly wrong with your run, it will never converge, most probably.
please remove all possible remnants from previous runs.