Weird Energy difference in two runs, bug in LDA?

[jzzhao]

hello:
I use two INCARs which have little difference to calculate BiCrO3. use GGA, the energy difference is neglectable, but use LDA or LDA+U, the energy difference almost 0.3 eV. I checked external pressure in two runs, they both not more than 0.05 KB, follow are my some test infos and input file,
I'm very appreciate for any comment and advise!

###########INCAR A################
LREAL= Auto
ISTART = 0
ICHARG = 2

ENCUT = 500
ENAUG = 606

ISYM = -1
ISIF = 3
IBRION = 2
NSW = 200

ISMEAR = -5
ALGO = Fast

PREC = High
EDIFF = .1E-5
EDIFFG= -0.005
#VOSKOWN = 1

ISPIN = 2
#LSORBIT = T
#SAXIS=0 0 1
MAGMOM = 12*0 4 4 -4 -4 4*0

LMAXMIX=4
LORBIT=11
LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = -1 2 -1
LDAUU = 0.00 3.00 0.00
LDAUJ = 0.00 0.8 0.00
LDAUPRINT = 2
LWAVE = F

##########INCAR B#################
LREAL= Auto
ISTART = 0
ICHARG = 1
#NEDOS=1001

ENCUT = 500
#ENAUG = 606

ISIF = 3
IBRION = 2
NSW = 100
EDIFFG = -0.005

ISMEAR = 0
ALGO = Fast

PREC = High
EDIFF = .1E-5

ISPIN = 2
#LSORBIT = T
#SAXIS=0 0 1
MAGMOM = 12*0 -4 -4 4 4 4*0

ISYM = -1
#LORBMOM = T
VOSKOWN = 1

LMAXMIX=4
LORBIT=11
LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = -1 2 -1
LDAUU = 0.00 3.00 0.00
LDAUJ = 0.00 0.8 0.00
LDAUPRINT = 2
LWAVE = F




Difference of INCAR A and B
< #ICHARG = 1
< #NEDOS=1001
---
> ICHARG = 2
7c6
< #ENAUG = 606 ! default for ENAUG is 605.4
---
> ENAUG = 606
10,12c10,11
< IBRION = 1
< NSW = 100
---
> IBRION = 2
> NSW = 200
14c13
< ISMEAR = 0
---
> ISMEAR = -5
> #VOSKOWN = 1
23,27c24
< MAGMOM = 12*0 -4 -4 4 4 4*0
< #LORBMOM = T
< VOSKOWN = 1
---
> MAGMOM = 12*0 4 4 -4 -4 4*0

###############KPOINT###########
auto
0
Gamma
7 7 4
0 0 0
###############POSCAR###############
O Cr Bi
1.00000000000000
4.4438921408257039 -2.6954963609271609 -1.4858763710071576
4.4438718378774436 2.6954685071796103 -1.4858728662434544
-0.0014575043115403 0.0000118507533469 9.4844099584575261
12 4 4
Selective dynamics
Direct
0.8787891342427829 0.2915046781566215 0.5853693199170066 T T T
0.6313255584383914 0.6748026490568246 0.3634037672512013 T T T
0.8351375656415144 0.8714378497991977 0.1566530485944131 T T T
0.7084903538831636 0.1212112507335350 0.9146325689308696 T T T
0.3252026555151300 0.3686735828988399 0.1365960895956527 T T T
0.1285630589902744 0.1648631078186231 0.3433470775779519 T T T
0.1212115227923961 0.7084974558334448 0.4146296390835303 T T T
0.3686686695821068 0.3252009416457144 0.6365991779787852 T T T
0.1648638341121057 0.1285710049845429 0.8433387657724766 T T T
0.2915044358946245 0.8787876256610941 0.0853703697354701 T T T
0.6747903112801826 0.6313248411578385 0.8634015533745384 T T T
0.8714486347699240 0.8351204427032666 0.6566557637550351 T T T
0.7583760714214348 0.2416355159687649 0.7500005469735528 T T T
0.2416256339135830 0.7583725566885684 0.2499989653732522 T T T
0.0000074542119899 0.4999946607212294 0.4999958534776550 T T T
0.4999896869737006 0.0000003972617794 0.0000060847938804 T T T
0.9102268955677152 0.3541066491757205 0.1305543613525420 T T T
0.6458906024809151 0.0897716126359320 0.3694437702051915 T T T
0.0897735514897732 0.6458962937557873 0.8694423497487217 T T T
0.3541143687983127 0.9102268833426749 0.6305609265082806 T T T




I do some test,
INCAR A， kpt 9 9 6
F= -.15494951E+03 E0= -.15494951E+03 d E =-.154950E+03 mag= 0.0000
INCAR A kpt 7 7 4
F= -.15494946E+03 E0= -.15494946E+03 d E =-.154949E+03 mag= 0.0000
INCAR A ENCUT=700, kpt 7 7 4
F= -.15502950E+03 E0= -.15502950E+03 d E =-.338192E-05 mag= 0.0000
INCAR A vasp5.2 7 7 4
F= -.15494946E+03 E0= -.15494946E+03 d E =-.154949E+03 mag= 0.0000

INCAR b kpt 9 9 6
F= -.15525955E+03 E0= -.15525955E+03 d E =-.380219E-04 mag= 0.0000
INCAR b kpt 7 7 4
F= -.15525955E+03 E0= -.15525954E+03 d E =0.414750E-05 mag= 0.0000
INCAR b ENCUT=700, kpt 7 7 4
F= -.15533974E+03 E0= -.15533974E+03 d E =-.658052E-06 mag= 0.0000
INCAR b vasp5.2 7 7 4
F= -.15525988E+03 E0= -.15525987E+03 d E =-.155260E+03 mag= 0.0000
INCAR b 7 7 4 change IBRION to 1
F= -.15525948E+03 E0= -.15525948E+03 d E =-.778590E-04 mag= 0.0000
INCAR b 7 7 4 change ENAUG to 606
F= -0.155259372627E+03 -0.50169E-05 -0.12762E-04111643 0.466E-02 0.404E-03
INCAR b 7 7 4 change ISMEAR to -5
F=-0.155259343036E+03 0.39816E-04 -0.14618E-03137145 0.157E-01 0.218E-02

Some of above calculations haven’t got convergence for we can already see the tendency, For other small systems I find use VOSKOWN = 1 or not have no effect in total energy. May be all these little difference amounts to the final large discrepancy?
We use the -154.9 INCAR to calculate the -155.2 final structure, surprisingly, the total energy converge to -154.9 again, seems that we can’t say the -155.2 structure is the global minimal.

For another LDA+U fully relaxed structure, we also use the above two INCAR
INCAR A
F= -.15519292E+03 E0= -.15519292E+03 d E =-.282901E-04 mag= 0.0000
INCAR B
F= -.15519312E+03 E0= -.15519312E+03 d E =-.155193E+03 mag= 0.0000
This case there is only 0.02meV difference.

is this a bug in LDA and LDA+U?

[admin]

if you use different BZ integration schemes, one applying a smearing on the levels (ISMEAR=0) and one without (tetrahedron methods) you should nor wonder about differences. Furthermore, please check if the k-mesh you use for the tetrahedron method really covers the IBZ completely

[boris]

Hi

There are a lot of metastable states in DFT+U that complexify (a lot) the convergence to the ground state. It's a well known problem with the DFT+U approximation and it's probably the reason why you obtain different states with the two INCAR files.

Some groups have worked on how to ensure the ground state is reached, you can find it on the Internet.

Regards