Hints on procedure: single molecule spectra

[andrea marini]

But even after uploading my public keys and waiting some time, it doesn't accept it.

Dear Bob, I think it takes some time to update the public keys uploaded on the server. In case the problem will be not fixed by tomorrow morning post again here and I will contact the administrator.

[bob]

OK, now it worked. Looks like one has to wait longer than one hour until the keys are functioning...

BTW, does anybody by chance have an idea how to compile NetCDF on the Power6 machine in Garching? I had a look at other posts on that topic but nothing worked for me so far...?

[bob]

Hi people,

after compiling the SVN version of yambo, things seem to run smoothly.

Now, I wanted to calculate the QP corrections to the energy levels of my molecule. Since there is no tutorial available about this topic, come back here for some assistance.

Here are more details:
The LDA calculation supplied 100 energy levels, of which 76 are occupied. So far, I included all 100 bands in the calculation of the polarization function (which may be a bit of an overkill?). I then tried to test, how much the QP corrections depend on the response block size by increasing NGsBlkXp. "How much" convergence can be expected? I increased NGsBlkXp up to 5000 but the corrections for [HOMO-4:LUMO+4] still change by ~0.1-0.2 eV. I tried to run 20000 on a 32 node machine, but the RAM requirements exceed the limits of the machine.

I would now go back and reduce the number of bands for the polarization function at a smaller NGsBlkXp, to see how that affects the calculated values. Does anyone here have any experience with such single molecule systems and knows a rough estimate of the parameter space I have to explore?

Cheers,
Bjoern

[Daniele Varsano]

Dear Bjoern,
it looks that you have a quite big molecule. I do not want to scare you but if you have 76 occupied bands,
I think you will need much more that 100 bands. Note that the number of bands have to converge both in the
response for the plasmon pole approximation (BndsRnXp) and also in the GW summation (GbndRnge).
Usually it is a quite slow convergence. Anyway, if you are then interested in optics, you can look at energy differences more than absolute values. Usually these converge much faster. Anyway regarding NGsBlkXp I would say that 20000Gblk is really
too much. ABout memory issues, you can try to consider to use a lower number of FFTGvecs, usually you do not need
all the Gvec you used for the ground state, and you can check in the report the orthonormality of the wave functions.

In summary, the parameter that have to be converged are:

EXXRLvcs= RL # [XX] Exchange RL components
BndsRnXp
NGsBlkXp
GbndRnge

Then the Plasmon Pole imaginary energy PPAPntXp, should also be checked, as the plasmon is not a well defined entity for finite systems, even if it usually works.

Another important issue are the volume size effects, these could be very important, and you may have to use cutoff coulomb techniques, you can have a look at Phys. Rev. B 73, 205119 (2006) that you can find here. In order to activate it, type yambo -c .

Hope it helps,

Daniele

[bob]

Dear Daniele,

Dear Bjoern,
it looks that you have a quite big molecule. I do not want to scare you but if you have 76 occupied bands,
I think you will need much more that 100 bands.

It is indeed not a particularly small molecule. Eventually, we will have a look at even bigger systems, and I know that this will be very challenging...

Note that the number of bands have to converge both in the
response for the plasmon pole approximation (BndsRnXp) and also in the GW summation (GbndRnge).
Usually it is a quite slow convergence. Anyway, if you are then interested in optics, you can look at energy differences more than absolute values. Usually these converge much faster. Anyway regarding NGsBlkXp I would say that 20000Gblk is really
too much.

Will do so. The thing with NGsBlkXp=20000 was that I did a run with 5000 on a single CPU on my office desktop and then switched to the big cluster machine and thought I can shoot with this big cannon at the molecule. Apparently, such a calculation would require more than 64GB of memory, if the LoadLeveler on that thing is to be believed.

ABout memory issues, you can try to consider to use a lower number of FFTGvecs, usually you do not need
all the Gvec you used for the ground state, and you can check in the report the orthonormality of the wave functions.

In summary, the parameter that have to be converged are:

EXXRLvcs= RL # [XX] Exchange RL components
BndsRnXp
NGsBlkXp
GbndRnge

Then the Plasmon Pole imaginary energy PPAPntXp, should also be checked, as the plasmon is not a well defined entity for finite systems, even if it usually works.

I saw that in the SVN version, there is an option to do real-axis GW. Would that, in principle, be a more suitable approach for a finite system?

Another important issue are the volume size effects, these could be very important, and you may have to use cutoff coulomb techniques, you can have a look at Phys. Rev. B 73, 205119 (2006) that you can find here. In order to activate it, type yambo -c .

Funny, I knew that paper because I implemented this cutoff technique in my own (now defunct) DFT code a couple of years ago, when I was trying to look at a charged system. I wonder how the volume size effect is affecting my present calculations? Is it due to the long-range nature of corrections?

Anyway, thanks a lot of all the help so far! I greatly appreciate the input!

Cheers,
Bjoern

[bob]

OK, here I am with more questions after I have now obtained an optical spectrum from the BSE.

1) I wonder how I can analyze the peaks, e.g. in terms of the coupled transitions. Can I look at lifetimes of excitations?

2) Can I differentiate between singlet and triplet excitons (which are not in the optical spectrum)?

Cheers,
Bjoern

[myrta gruning]

Dear Bjoern

The possibility of analyzing the spectrum in terms of KS transitions and of obtaining zero oscillator strength excitations (e.g. triplet) depends on the method for solving the BSE and of the inclusion of a key in the input to write away the excitons WFs

If you used the haydock solver
BSSmod= "h"
all you can obtain is the spectrum. To get the info you wish you should use the diagonalization
BSSmod= "d"
and include the keyword
WRbsWF

The you can use ypp to analyze the excitations.
ypp -e <opt> to generate the input
use ypp -H for the <opt>
Also look at the yambo reference paper for more details on the BSE solvers.

Regards,
m

[Daniele Varsano]

Dear Bjoern,

In order to look at the KS transition of the excitations, you have to use the ypp utility as suggested by Myrta. In this way you can identify the dark transition, which have small or zero oscillator strength.
About the triplet, in order to calculate them, you have to construct the BSE matrix in different way (omitting the x part), please have a look at this paper:
M. Rohlfing and S. G. Louie
Electron-hole excitations and optical spectra from first principles
Phys. Rev. B 62, 4927 (2000).

Hope it helps:

Bests,

Daniele

[bob]

Hi Daniele,

In order to look at the KS transition of the excitations, you have to use the ypp utility as suggested by Myrta. In this way you can identify the dark transition, which have small or zero oscillator strength.

Is there any detailed documentation on using ypp?

About the triplet, in order to calculate them, you have to construct the BSE matrix in different way (omitting the x part), please have a look at this paper:

OK, I reran the calculation with

Code: Select all

BSresKmod= "c"              # [BSK] Resonant Kernel mode. (`x`;`c`;`d`)
BScplKmod= "c"            # [BSK] Coupling Kernel mode. (`x`;`c`;`d`;`u`)

Then there is another peak showing up in Im(eps) at lower energy. I thought, however, that the triplet should not show up in the absorption?

[myrta gruning]

Hallo

I am not aware/do not know about the ref Daniele gave you, but when you omit the x you are omitting physical effects, in particular local fields. Local fields may both blue shift or change the strength of excitations.
You can also plot the independent particle spectra to have an idea of the various effects. It should be on the 4th and 5th column (look into the file to be sure)

About ypp there is no documentation, but using the command line
ypp -e <opt>
for generating the input is quite self-explaining, moreover the runs are fast (you can run it on your PC, is just post processing data).
You should start by sorting the excitation (opt=s), so that you know the numbers associated to each excitation (that you need to specify if you want e.g. to get the excitonic WF). Then you can analyze the amplitude (a option, this gives you the composition of the excitations in terms of KS transitions) or plot the exciton wave function (w option).

regards,
m