RIM convergence

[Laura Caputo]

Hello everyone,

I have a question concerning RIM values convergence. I am studying a 2D material and want to calculate GW band gap. To use the RIM approach, I have been studying the convergence of both RandQpts and RandGvec with the HF band gap before moving to the GW gap study.
However, when I enhancing the values of these, the gap does not convergence smoothly.
I noticed that RL volumes are however very similar:
RIM RL volume [a.u.] : 0.027870
Real RL volume [a.u.] : 0.027866
So, my question is: how can I check the convergence of these parameters?
In attachment a report file.

[Daniele Varsano]

Dear Laura,

the gap is changing of few meV, I would say less than that the accuracy of the method.
Note that RIM is a stochastic integration, so you have a statistical error associated, if you consider more random Q points (let's say 3 or 5 millions) you will have a smoother curve, I presume.

Please note that in the report you included the RIM it is not activated.
Note also that for 2D systems the use of truncated coulomb cutoff is encouraged and the cell should be large in the non periodic direction in order to isolate the system.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

I have considered from 1 to 7 million, and I have the graph reported. However, as you said, if I have a few meV differences, I can consider this converged, I presume.

Regarding, the RIM not activated: I included the 'rim_cut' line and the RandQpts+RandGvec variables. Do I need to add anything else? I was following the tutorials and these are the only ones listed.

Lastly, thanks for the tip about 2D system. However, in that case, I have read that a slightly smaller value of the unit cell on the non-periodic axis is needed. For example, if I have 10Angstrom vacuum along z, is 9.9 good for the Coulomb potential calculation to be considered in the yambo calculation? Moreover, do I need to move the .cutoff database to the SAVE folder?

Thanks,
Laura

[Daniele Varsano]

Dear Laura,

looking at your report rim_cut is not included (see the mirror of the input file at the end), and when reading the HF database I can see:

Code: Select all

 RIM RL components                                : 0
 RIM random points                                : 0

if this is the case the oscillation you are seeing are due to the single precision of the compilation, but again they are totally fine.

I have 10Angstrom vacuum along z, is 9.9 good for the Coulomb potential calculation to be considered in the yambo calculation?

Yes it is! depending on the extension of your system (charge) in the non periodic correction 10 Angstrom could be not optimal, but you can have a try.

Moreover, do I need to move the .cutoff database to the SAVE folder?

As all the other databases, Yambo reads all the databases present in the SAVE and in the directory indicated by -J. If -J is not used all the databases are placed in the SAVE directory by default. If you move it in the SAVE it sill be read in every run independently of the -J label.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

Sorry for the mistake, I uploaded another file without noticing.
In the attachment the correct one that I used. Sorry again.

Regarding the z-axis value, I have done different DFT calculations on these systems with a 10Angstrom size without having any problems. Do you suggest adding some Angstrom in GW calculations?

Sincerely,
Laura Caputo

[Daniele Varsano]

Dear Laura,

Do you suggest adding some Angstrom in GW calculations?

No, just to inform that is always a good practice to check.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

Thanks for all the clarifications and support for new Yambo users.

With regards,
Laura Caputo

[Laura Caputo]

Dear Daniele,
As you suggested, I have tried to use a truncated potential for my system. However, the GW gap differs for 0.6eV in between using or not the truncated potential (maintaining the RIM activated). I used a box size slightly smaller than my cell. I attached my report file (only RIM activated) and the report_box file (RIM+box).

Moreover, I understand that the use of RIM is mandatory for numeric stability for 2D materials. Is the truncated potential encouraged for faster convergence tests or is it mandatory for low dimensional materials?

[Daniele Varsano]

Dear Laura,

GW calculation converge very slowly with respect the size of vacuum, this is why the truncated coulomb potential needs to be used. Then in order to build the cutoff box, also the RIM is needed.

Anyway there is a problem in your calculation:
it seems it has been done on top of a previous calculation with different parameter where you used a box of side 9.90 which is not realistic as it is very small compared with your cell.
As you can see in the report you have many Warnings:

Code: Select all

[RD./gw_box//ndb.pp]------------------------------------------------------------
  Brillouin Zone Q/K grids (IBZ/BZ)                :   13   36   13   36
  RL vectors                                       :  14769 [WF]
  *WRN* Coulomb cutoff potential                   : box z  9.900

you are using the screening calculated with that box values. The same is true for the HF part, and finally the QP energies are not calculated but read fro the previous database:

Code: Select all

[RD./gw_box//ndb.QP]------------------------------------------------------------
  Lattice constants                                :  13.99321  12.11848  18.89726
  *WRN* Coulomb cutoff potential                   : box z  9.900

RD stands for read, while new calculation are performed you will see a WR in correspondence of the databases.

So you should repeat the calculations deleting the previous databases or using a different label.
Best,
Daniele

[Laura Caputo]

Dear Daniele,

I didn't notice it, sorry I'm learning to properly use Yambo and read the reports and thus I totally missed it. I'm re-running it.

For the usage of the truncated potential, in the tutorial, I read that we have just to set the value of the z in the box. Do I need to run some convergence for this value, or it's fine to just select a slightly smaller value than the 'original' cell?

Sincerely,
Laura