VASP Forum

Dear All,

I would like to use TDDFT+U to obtain the optical absorption of my cluster system using ALGO = TDHF and TIMEEV, following the procedure below:
1- Perform a ground-state calculation with LOPTICS = .TRUE. and include empty bands with DFT+U.
2- Calculate the optical absorption using TDHF and TIMEEV.

#TDHF

DFT+U parameters 
ALGO = TDHF
LHARTREE  = .TRUE.
LADDER    = .FALSE.
LFXC      = .TRUE.
DIM = 0
IBSE = 0
ISYM = 0

#TIMEEV 

DFT+U parameters 
ALGO = TIMEEV
NELM   = 20000
LHARTREE  = .FALSE.
LADDER    = .FALSE.
LFXC      = .TRUE.
DIM = 0
ISYM = 0
IEPSILON = 4

However, I have a few questions:
1- I am not sure whether LHARTREE should be set to TRUE or FALSE in the second step. When LHARTREE = FALSE, TDHF gives spectra similar to IP by LOPTICS, while in TIMEEV there seems to be no difference compared to TRUE. Could you please clarify this behavior? Should I set LHARTREE = TRUE for TDHF and FALSE for TIMEEV?

2- I observed that the absorption spectrum calculated with TIMEEV is red shifted compared to that obtained with TDHF. What could be the reason for this shift?

3- In the OUTCAR file for TIMEEV, I obtained the following output: maximum number of steps NELM    =   20000 number of time steps:  1176,  what does it mean? I suppose to have 20000 steps.

4- In this 0D structure with DIM = 0 in the INCAR file, there seems to be no difference compared to the default DIM setting of 3. Does the calculated dielectric function depend on the vacuum size? If so, does that mean that quantities such as the polarizability and absorption coefficient cannot be calculated using their conventional formulas for bulk systems based on the calculated dielectric function?

Many thanks in advance,
Ibrahim

Dear Ibrahim,

Thank you for describing your approach.

add 1) the choice of LHARTREE depends on the desired level of approximation:

LHARTREE=T; LADDER=T ! full TDHF/TIMEEV
LHARTREE=F; LADDER=T ! only ladders (exitonic effects)
LHARTREE=T; LADDER=F ! only bubbles (RPA)
LHARTREE=F; LADDER=F ! IPA

To compare the two methods, you probably want to use the same level of approximation for both.

add 2) I am afraid that the interpretation of the data is beyond the scope. But we can make sure together that the result you are seeing is indeed what the method yields and not an artifact due to a problem in the setup.

add 3) with the information you shared, I assume you may need to consider the effect of the tags CSHIFT and OMEGAMAX. See setting up the time step. If that does not resolve the qustion, please share a minimal example so we can provide a proper more complete answer.

add 4) I will ask what is the status of this tag. It is not documented, so I am not sure it is supported at all.

Generally, I am not an expert in these calculations, but I will ask my colleagues. Meanwhile, I assume you are aware of the following two how-to pages on the Wiki:

• Time-evolution algorithm

• Time-dependent density-functional theory calculations

Does this help already?

Dear marie-therese huebsch,

Thank you for the reply.

1) I think that, in addition to the choice of LHARTREE and LADDER, the LFXC tag may also be mentioned on the VASP wiki page (https://www.vasp.at/wiki/LHARTREE). It would also be useful to clarify whether different combinations of these three tags lead to distinct approximations in the cases of TDHF or TIMEEV. 
- According to the VASP wiki page on the time-evolution algorithm (https://vasp.at/wiki/Time-evolution_algorithm), "the default setup in VASP is LHARTREE = .FALSE., LADDER = .FALSE., and LFXC = .TRUE.. This means that, if no tags are specified in the INCAR file, the time-propagation run will use the TDDFT kernel". However, based also on VASP manual, the default for LHARTREE is True, not False and LFXC is False not True. I hope you can check this wiki page.
- Also, with TDHF and LHARTREE = .FALSE., LADDER = .FALSE., LFXC = .TRUE., I obtained results similar to those from IP by LOPTICS. While with TIMEEV there seems to be no difference between (LHARTREE = .FALSE., LADDER = .FALSE., LFXC = .TRUE.) and (LHARTREE = .TRUE., LADDER = .FALSE., LFXC = .TRUE.).

2) Should I expect different behavior between TDHF and TIMEEV for the same system?

3) In OUTCAR, I would like to understand what is the meaning of " maximum number of steps NELM    =   20000" and "number of time steps:  1176"?

4) Yes, DIM is not documented, however, I found it in OUTCAR so I included it in INCAR. And asked whether the calculated dielectric function depends on the vacuum size, in order to account for other quantities such as the polarizability and the absorption coefficient.

Many thanks in advance,
Ibrahim

Dear Ibrahim,

I will try to give some help on this issue.

The first point I would like to clarify is that both TIMEEV and TDHF should give the same result, provided that the inputs are properly set. This is because both can work at the TD-DFT level, with TDHF relying on the Casida equation, while TIMEEV performs an actual time-propagation algorithm and then uses a Fourier transform to compute the frequency-dependent dielectric. I admit that the names on both pages are not the best.

The second point is that using LADDER, LHARTREE, and LFXC depends on which system you are studying. For smaller molecules you should be fine with just LHARTREE=.TRUE. and LFXC=.TRUE. (so basically TD-DFT), while for larger systems you need a good description of the long-range interaction between electrons and holes, so LADDER should be set to .TRUE..

From your inputs however, it seems that you are not setting up the parameters needed for the exchange-correlation kernel, namely AEXX. You need this tag to specify the amount of exchange in the kernel.

My last point, for now, is what kind of exchange-correlation functional are you using? The TDHF algorithm is in principle faster, because the TIMEEV requires the density and potential to be updated at every time step.

Kind regards,
Pedro

Dear Pedro,

Thank you for your detailed reply.

I am using TDDFT+U with PBE to approximate the exchange–correlation functional, so I set LADDER = .FALSE.. However, I am not sure whether LHARTREE should be set to .TRUE. or .FALSE.in this case?

I am not using a hybrid functional, so AEXX is zero in my case.

Also, for a finite system in vacuum using VASP, does the calculated dielectric function depend on the size of the vacuum region? If so, is it possible to obtain the absolute dielectric function of the cluster itself, in order to accurately determine related quantities such as the polarizability or the absorption coefficient?

Many thanks in advance,
Ibrahim

Dear Ibrahim,

From what you wrote, yes, you should set LHARTREE=.TRUE. in the INCAR.

Regarding the issue of volume, yes, the dielectric function that is written by VASP uses the full volume of the cell you provide in the POSCAR, so the vacuum is included. If your system has lower dimensionality you will need to account for this.

Kind regards,
Pedro