How to restart DOS calculations with MBJ+SOC

[rishikanta_m]

Hi

I run a calculation for density of states using MBJ+SOC and finally, nearly at the end, just after the convergence, it crashes (may be because of memory requirement for writing WAVECHAR).

I would like to restart the calculation, since it already took a lot of time. Please help me.

[ferenc_karsai]

What do you mean by restart?
If you run out of memory, possibly it will happen again on the same machine. So in that case you need to lower the accuracy of some parameters or go to a bigger machine.

[fabien_tran1]

Hi,

With MBJ it is unfortunately not possible to restart a calculation without having to repeat the self-consistent-field procedure. Even if the WAVECAR file would be present, because of the calculation of the parameter c in the MBJ (however if c is set to some fixed value with the CMBJ tag in INCAR, then restarting from a WAVECAR works).

Making possible to restart a MBJ calculation (with or without WAVECAR file) is on our todo list.

Concerning the crash at the end of the calculation: you can quickly test in advance if it occurs by setting NELM=1 (only one iteration is done) in INCAR.

[rishikanta_m]

Thank you for your suggestions. Will check it out.

Another thing I want to know in MBJ+SOC calculation:

Can we use the WAVECHAR generated in DOS calculation for Optical properties calculations (mbJ+SOC)?
If yes, would it help to reduce the computational time?

[fabien_tran1]

It may depend on the used procedure. Can you please describe the steps that you would use to calculate the optical properties with mBJ+SOC?

[rishikanta_m]

The procedure that I am considering is like this:

1. PBE+SOC-SCF calculation
2. MBJ+SOC-DOS calculation. (WAVECHAR from 1)
3. MBJ+SOC-Optical calculation. (WAVECHAR from 2)

My concern is about going from step 2 to step 3.

[fabien_tran1]

This corresponds to three different INCAR files, right? Can you provide them?

[rishikanta_m]

Yes, the input INCAR files are different.
I have attached the input files. Please correct me if there are any mistakes.

[fabien_tran1]

The INCAR for step 3 is not ok, because the MBJ and SOC blocks are missing (the calculation of optical properties needs of course to be done with the same methods as for the previous step). In addition, to restart a MBJ calculation from a WAVECAR you can use the trick which consists of fixing CMBJ to the value that was obtained at the previous step (step 2). Thus, the INCAR for step 3 should be:

***************************
ISTART = 1
ICHARG = 0
ENCUT = 400
ISMEAR = 0
NSW = 0
EDIFF = 1E-6
NELM = 300

#SOC-related:
LSORBIT=.TRUE.
SAXIS= 0 0 1
ISYM=0

#MBJ:
METAGGA=MBJ
CMBJ = XXXXXX
LASPH=.TRUE.
LMAXMIX=4

#Optical tags
ALGO = Exact
NBANDS = 80
LOPTICS = .TRUE.
CSHIFT = 0.100
NEDOS = 1000
ISMEAR = 0
SIGMA = 0.1
EDIFF = 1.E-6
***************************

where you have to replace XXXXXX by the value of CMBJ that was obtained at step 2 (search for CMBJ in OUTCAR or vasprun.xml).

[rishikanta_m]

Yes, I forgot to put those tags. (#SOC #MBJ).
Thank you so much, Sir. I got your point. Will implement your suggestions.

[rishikanta_m]

Regarding CMBJ value,

I have earlier done bandstructure calculation under the same methodology (mBJ+SOC), can I use the same CMBJ value for calculations of DOS and Optical Properties?

[rishikanta_m]

Another problem.

In Optical properties using MBJ+SOC, if I used ALGO=Exact. VASP crashes due to memory requirement.

So, I changed it to ALGO=Normal.
(According to forum/viewtopic.php?t=18131)
Near the end of the simulation, it says

"You have switched on METAGGA and linear response routines. |
| This combination is currently not supported. |
| If you have selected LOPTICS=.TRUE., VASP will continue but some |
| matrix elements [H,r] are neglected (you might try to use |
| LPEAD=.TRUE. which works irrespective of H). |"

What can be the possible workarounds?

[fabien_tran1]

This message is important for the traditional meta-GGA functionals like SCAN or TPSS, but can be ignored for MBJ which is a different type of meta-GGA (compared to a GGA, MBJ has no additional matrix element). That also means that setting LPEAD to .TRUE. or .FALSE. should in principle not matter with MBJ (but test it yourself to be sure).

[rishikanta_m]

Unfortunately, after this message, VASP does not write the real and imaginary dielectric function.
I have tried this calculation several times with ALGO=Normal (with SOC) changing NBANDS and setting/not setting CMBJ value.
Everytime, it runs similarly completing the SCF convergence and at the end, no dielectric function output.

But when I ran the MBJ calculation without SOC, I got the same message but it perfectly outputs the dielectric function.
Is there a problem with SOC? Please have a look at my files.

[fabien_tran1]

I ran your case with SOC but with less k-points (6x6x3) to make the calculation faster. After the warning "You have switched on METAGGA and linear-response ...", the dielectric function is printed. Maybe your SOC calculation crashed just after the warning because of the memory demand that is too large. Please, run again your SOC calculation with 6x6x3 to see if it works also for you.