RIM convergence

[Daniele Varsano]

Dear Laura,

the position of the system, in principle should not affect, as you say the Coulomb potential is a function of |r-r'|.
Anyway note that your calculation, due to the different position in space have a different number of symmetries.
Moreover it seems to me that something happened in the ground state calculation as you have different gap at gamma:

report_0

Code: Select all

 [X] Direct Gap                                    :  0.604562 [eV]

report-0.25

Code: Select all

 [X] Direct Gap                                    :  1.551236 [eV]

please check what happened to the band structure as it should not change in the two case.


The box cutoff, anyway i general has some problem of accuracy, and it will be soon dismissed and substituted with a more accurate method.
If you want in the meantime you can give a try to use the "ws cutoff":

Code: Select all

CUTGeo= "ws z"

this is the recommended technique when using orthorhombic cells. For non-orthorhombic cell it can fails, but you can have try, if there are problem the code will stop.

I have considered the k point sampling that I used for the optimization after a convergence study on the final energy and I am calculating the gap on Gamma. Is it good this way or should I do different Abinit runs with different k-points meshes to check the GW gap? The last option could take a lot of time.

Unfortunately, the k point convergence is something that has to be checked. GW band gap and absorption spectrum in BSE usually require grid larger than the one needed to converge the ground state density, but this depends of the bands profile, exciton localisation etc...

Best,
Daniele

[Laura Caputo]

Dear Daniele,

So, as I understand, the GW gap should not change if we translate the system along z in a rigid way.

For the CUTGeo= "ws z", is it more accurate? Should it give very similar results to the 'box' I used?

Lastly, for k-points convergence: I have done all the convergence tests for the bands etc using a 6x6x1 grid arriving at the convergence point for GW gap. Could I use these converged values to calculate the gap at different meshes to see the convergence?

Thanks,
Laura

[Daniele Varsano]

Dear Laura,

So, as I understand, the GW gap should not change if we translate the system along z in a rigid way.

Yes, but also the DFT gap should not, so I do not know what happened in your DFT calculation.

For the CUTGeo= "ws z", is it more accurate? Should it give very similar results to the 'box' I used?

I would say yes, you do not need to take care of the size, and RIM is not needed anymore, but it depends on the k points grid. But again I 'm not sure the code will not complain as you have a non orthorhombic cell. You should obtain a similar result.

Could I use these converged values to calculate the gap at different meshes to see the convergence?

Yes, as bands and gvectors are independent on the k point sampling and you can transfer them directly.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

Yes, I'm going to check what went wrong in the DFT calculation indeed.

I have a last question for you since I'm now reading something more about the ws box: since I have done different calculations with 'box z' + Rim, can I use the same converged parameters used for box z for the calculation with was cell? If the results do not change much, should it be good to leave the box z parameter?

Thanks,
Laura

[Daniele Varsano]

Dear Laura ,
sure you can go on with box cutoff coulomb potential. My suggestion was just in case you wanted to explore an alternative method, but you can go on with what you did. In any case, all the other parameter are independent, except for the k points sampling.
Best,
Daniele

[Laura Caputo]

Dear Daniele,

Thanks a lot for your replies and your constant support.

Have a nice day,
Laura Caputo

[Laura Caputo]

Dear Daniele,

Sorry for posting another question, I want to get all things right before moving further into running calculations.

Regarding convergence, the k-point should be converged. However, the plane wave cut off used for the DFT runs shouldn't be converged, right?

Moreover, regarding the EXXRLvcs and VXCRLvcs, I have seen on the tutorial that only the first is converged within HF while the second one is left at maximum. However, I will have particularly big system on which I will try to calculate the GW gap. Do you recommend to converge also the G-vectors of the density to make the calculation less demanding?

Thanks,
Laura

[Daniele Varsano]

Dear Laura,

However, the plane wave cut off used for the DFT runs shouldn't be converged, right?

Yambo uses the Gvector you used for the GS calculation. In general you can lower the FFTGvecs, controlling the orthonormalization of the states which is reported e.g. in an hf calculation.

VXCRLvcs is left to the maximum for safety reason, in LDA you can usually lower it, in GGA is more critical, you can anyway have a look to the value of the E_xc which is reported in the report file and compare with the one calculated by abinit, if it prints it. The two should be the same, unless you used pseudo potential with non linear core correction, which is anyway discouraged.

Please note that anyway that part of the calculation it is not demanding so you would not save much computational time.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

If I have understood correctly, after the convergence of the plane waves in DFT framework and thus their usage in the generation of the WFK file in abinit, these are automatically used by yambo by default and they should be enough (without converging the plane wave energy cut off in the DFT step vs GW gap, for example). Right?
In my case, I have set this parameter in default in my calculation, this should be safe, right?

Thanks for the clarification about VXCRLvcs.

Laura

[Daniele Varsano]

Dear Laura,
yes that's right, once the gs is converged than you apply perturbation theory on top.

Daniele