VXCRLvcs

[Laura Caputo]

Dear Daniele,

Sorry again for the question but I want to be sure it is 100% correct. Regarding EXXRLvcs, I have done the convergence test following the tutorial on the Yambo website. In that case, the convergence regarding the bang gap is carried out. In accordance to that, I have converged this parameter with the value of the HF gap. Is this correct or should I watch more specific parameters as in VXCRLvcs?

[Daniele Varsano]

Sorry Laura I do not understand your point:
EXXRLvcs: Number of g vector in the HF term
VXCRLvcs: Number of g vector in the xc term

In yambo they are indicated in the report as:
<i | nlXC | i> and <i|lXC|i> respectively (non-local and local).

convergence of HF is a standard procedure.
VXCRLvcs is usually kept as its maximum, but you can lower it. In this case you need to check you are not using a too small number affecting the final value.

This is not a calculation intensive part of a GW calculation, I personally, when possible I use to keep them at its maximum value so that I remove a convergence procedure and I am sure they are converged.

Best,
Daniele

[Laura Caputo]

Dear Daniele,

Thanks for the explanation.
I did as you explained for checking that VXCRLvcs is enough. I started from 107469RL (which was the one used by default by Yambo, that however didn't change with ecut) and decrease it to 106469, 105469 etc.

The E_xc value do not change, when I see the <n|lXC|n> value, I see this:
<36|lXC|36> = -14.77849 -0.38292E-22
<37|lXC|37> = -13.15058 -0.73576E-23

<36|lXC|36> = -14.77849 -0.28971E-22
<37|lXC|37> = -13.15059 0.35335E-23

<36|lXC|36> = -14.77846 0.17715E-22
<37|lXC|37> = -13.15055 -0.55994E-23

<36|lXC|36> = -14.77847 -0.33583E-22
<37|lXC|37> = -13.15055 -0.35032E-23

There are some low differences. Is this convergence enough?

Sorry again for the 'continous' answering!

[Daniele Varsano]

Dear Laura,

the change you are observing are much smaller than the precision one can expect for this kind of calculation, so it is more than ok.
Anyway in this way you are not gaining that much reducing the g vector of few percent.
Again, if you are able to perform calculations with all the gvectors, why you want to reduce them? This calculation are done once for all and stored in the ndb.HF* database.
I would concentrate in the more cumbersome convergence as bands summation and dielectric matrix size in the correlation self energy.
Maybe there is something misleading in the tutorial that brought you in this convergence study?

Best,
Daniele

[Laura Caputo]

Dear Daniele,

What brought me here is the fact that with different energy cut off in Abinit calculations, I obtained the same default value of VXCRLvcs.
I will explain a little bit better: a few days ago, I mistakenly calculated the ground state of a system with a 500eV cut off (very silly mistake, I did not notice that I didn't change this value). Since I didn't notice the error, I calculated the GW gap which was 2.88eV.

After noticing this mistake, I re-calculated the gs with 850eV cut off, which is the converged value. I noticed that, after converting the output files to yambo with the two different cut off, the default VXCRLvcs value suggested by yambo were the same (I checked that the cut off in Abinit where different). I calculated the GW gap using the same parameter values as in the previous calculation, obtaining 2.89eV GW gap, so very similar even if 500eV was not converged in gs.

In Abinit, unfortunately, I cannot find how to check the value of G vectors (I will ask in their forum). Since I had begun my calculations with their VXRLvcs, I wanted to make sure that this was enough.
I understand that it is good to left it as default by yambo, and this is what I usually do. However, after this problem, I wanted to be sure that my parameters were correct.