Different results for SOC with different versions?

[jhyoung]

Dear VASPers,

I have been trying to calculate the energy difference due to SOC for a given system.

When I computed the total energy with SOC included, the results with vasp.5.2 are different from those with vasp.5.4:
vasp.5.2 says that there is large SOC effect (energy difference with and without SOC is about 100meV).
vasp.5.4 says that there is no such energy difference (less than a few meV).
(In both cases (i.e. vasp.5.2 and vasp.5.4), INCAR, POSCAR, POTCAR and KPOINTS are exactly the same)

Without SOC, the two versions produced the same total energies.
I am wondering whether there are known issues regarding this (or, did I just wrongfully compile VASP?)

Thank you all,

Best regards,

[mmarsman]

Dear jhyoung,

There is a difference in the implementation of the SOC operator in VASP.5.2 and VASP.5.4.
To be more precise: a bug was fixed in the implementation of the SOC operator in VASP.5.4.1.05Feb16.
This fix was published in August 2016 as a patch to the aforementioned version (patch.5.4.1.03082016).
If you're interested to see the details of this patch you can download it from our vasp.5.X.X installation page on our wiki.

This in itself explains that you will find a difference in the total energies of VASP.5.2 and VASP.5.4(.4?) when SOC is taken into account.
On the other hand, though, an effect of 100 meV on the total energy seems too large to me to be attributable to this bugfix.
Of course one should factor in the size of your system. Is this a calculation on a large system? How many meV/atom is the difference?

If this is not a "large" system, then in all probability you have not converged to the same solution as before.
Did you analyse the solutions (local magnetic moments etc) you obtained to make sure they are what you think they should be?

Maybe you can provide some more details on what exactly you are calculating, then hopefully we can figure this out!

Cheers

[jhyoung]

Dear mmarsman,

Thanks for your kind reply.
Below are some details about my SOC calculation for your reference.

=== INCAR ===
SYSTEM = CH
NSW = 0
NELM = 500

IBRION = 2
LREAL = A
ENCUT = 550 eV
ENAUG = 1100 eV
EDIFF = 0.1E-4

ISMEAR = 0
SIGMA = 0.05
NEDOS = 2700

MAGMOM = 3*0 2*0 1 15*0 2*0 -1 6*0
LWAVE = .F.
LSORBIT = .T.


=== POSCAR ===
CH
1.00000000000000
0.0000000000000000 -2.4718333453518171 0.0000000000000000
8.5747774854717331 0.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000 20.0000000000000000
C H
8 2
Direct
0.000000 0.0059658038938977 0.4656704313077332
0.500000 0.2523616547070396 0.4402287253726978
0.000000 0.5009285862569968 0.4486505026458137
0.500000 0.7502993171216401 0.4569462046861292
0.500000 0.0773675294394380 0.4343295686922725
0.000000 0.3330340162116958 0.4430537953138765
0.500000 0.5824047470763389 0.4513494973541920
0.000000 0.8309716786262962 0.4597712746273079
0.000000 0.0379219475772266 0.5193850960421870
0.500000 0.0454113857561091 0.3806149039578121


=== KPOINTS ===
57x21x1
0
M
57 21 1
0. 0. 0.


=== POTCAR (only header is included because it is too long) ===
PAW_GGA C 05Jan2001
PAW_GGA H 07Jul1998


Comparison has been made between vasp.5.2.2 and vasp.5.4.4 with exactly same files.
While I am still struggling to find out what is going on, I hope the above info will help you figure out what the problem is.


Best regards,