Yambo Community Forum

Dear Laura,

Thanks. So, if I understand correctly, I should always calculate the QP corrections for all the k-points (since BSE uses all of them) and for the bands, the QP corrections of the bands used in BSEbands.

This is option 1 in my previous post. Then there is also option 2 (if they are not found yambo performs a fit), and eventually option 3.

BSE needs the static screening:
1) it can be read from the ndb.ppa as it is already contained in the ppa database (when building the input file you need to use the -X p option instead of -X s) The corresponding variables are BndsRnXp, NGsBlkXp etc... the runlevel is indicated as ppa em1d in input (em1s not needed).

2) Alternatively, if you do not have the ndb.pp files available, you calculate the static screening only, -X s option, em1s runlevel, BndsRnXs, NGsBlkXs... variables.

You do not need to change variables by hand if you provide the correct option when building inputs.

Best,
Daniele

Hello Daniele,

After some tries, I obtain the same problem. Please find attachment a couple of reports and output for different values of exx and Blk.

Dear Laura,

the system seems to converge wrt BSENGexx and BSENGBlk.
I would check convergence wrt BSEBands and also wrt the vacuum, despite you are using a coulomb cutoff the supercell size you are using is rather small along z.
As stated in the first post, the meaningful quantity is apha and not eps, in any case the two are proportional.


Best,
Daniele

Hello Daniele,

Thanks for your reply. So, I could use these very small values of exx and Blk? Because it seems that the peak is converged also at smaller values than these. Can I ask you what can cause the system to converge so fast against these values? For the vacuum, I have a monolayer and some time ago I have checked the convergence wrt the QP gap. Could it be different for BSE? This is why I asked some time ago.

Dear Laura,
if you have already checked the amount of vacuum for GW, it should be ok also for the BSE.
About the fast convergences, I cannot say much as I do not know the system you are dealing with.
About the convergences, it is very important to check the k point samplings, you are using a very small sampling and this affects the localization of the exciton.

Best,
Daniele

Dear Daniele,
I will try to do the same convergences with a larger mesh then, if it does not change I understand from your answer that in the report file nothing is strange/unread and I could proceed with the calculations.

Thank you,

I have just one last thing: if with larger meshes I find the same convergences, which is very very low, sometimes converged also at 1Ry for the exchange and correlation cut off energies, is it same to run calculations with such low values?

Dear Lura,
k point convergence is rather independent of the other parameters.
I suggest you anyway to increase the number of bands in the BSE.
Honestly, 1Ry sounds very a very low value but from your previous tests it seems, that's the case.

Best,
Daniele

Dear Daniele,

Do you mean that I should check these convergences by taking into account more bands? As you can see from some graphs in the previous messages, I tried to enhance little by little the bands and it's the only parameter that it seems to be converging 'normally'.

Moreover, is there any recommendation on how many bands to consider when doing these study convergences to have a big enough matrix?

Dear Laura,

Moreover, is there any recommendation on how many bands to consider when doing these study convergences to have a big enough matrix?

This is a parameter that needs to be converged looking at the range of excitations you are interested in. The small size of the matrix is mainly due to the very small k point grid you are using.

Also what seems a bit strange is your cell, where the y-direction is larger than the z-direction where it is supposed to be the presence of the vacuum.

The first inspection I would do is the convergence wrt the k points and eventually repeat your convergences tests using a larger grid in case your spectrum change qualitatively.

Best,
Daniele