TURBOMOLE Modules => Escf and Egrad => Topic started by: martijn on July 31, 2020, 01:53:31 PM

Title: BSE excitation calculations
Post by: martijn on July 31, 2020, 01:53:31 PM

I have two questions regarding BSE excitation calculations:

1) When I run a BSE calculation in 7.4.1 using say 3 roots by adding the following lines to the control file after running a GW calculation:

$scfinstab rpas
$soes all         3

and removing the $gw section, escf seems to calculates all possible (3000+) roots instead if the 3 I ask for. Similarly, when I specify each irrep seperately:

$scfinstab rpas
 a   3
 b1 3
 b2 3
 b3 3

I'm pretty sure that is not the expected behavior but I have no clue what I'm doing wrong.

2) How do the excitation energies calculated using BSE depend on the choices made when running the underlying GW calculation, especially if that GW calculation was done using ri-ac-gw or ri-cd-gw?

Naively it seems to me that calculation of excitations other than HOMO -> LUMO (for ri-ac-gw) or excitations involving other QPs than those included in the contour deformation (ri-cd-gw) could/should be problematic. However, even for those excitations, the inverse of the dielectric function used to screen the exchange, depends, if I'm not mistaken, on all quasiparticle energies.


Title: Re: BSE excitation calculations
Post by: chris.hol on August 01, 2020, 11:26:31 AM

1) Actually I fail to reproduce this, the section you specified calculates 3 roots each for me :/ . May it be that there is another $soes group left in your input? (Or something else interfering)

2) Indeed, there is an dependence on the choice of the GW calculation, and therefore to the approximation to the quasiparticle energies. If lets say HOMO and LUMO contribute to a BSE excitation than the difference between ri-ac-gw and ri-cd-gw will break down to the (usually small) difference between the quasiparticle HOMO and LUMO. If more quasiparticle states are involved things get more involved of course.

Further, as you stated correctly the dielectric function depends on all quasiparticle energies, therefore in the best case all of them are available - but for large systems this is barely an option. Still excitation that mainly happen between orbitals were the quasiparticle energies have been determined from ri-cd-gw should be fine, as the main error stems from the zeroth-order quasiparticle gap and not the dielectric function in my experience. But the actual experience on the exact influence of the different GW approximations on the BSE excitation energies is limited. As the GW calculation is currently the limiting factor in the accuray of a BSE calculation some research on this is done on improving also the "uncorrected" quasiparticle states AFAIK, though any experience/additional research and knowledge in this respect would be very welcome.

All the best,
Title: Re: BSE excitation calculations
Post by: martijn on August 02, 2020, 01:57:23 AM
Thanks Christof! Regarding (1) I was an idiot. Yes, you're correct. I had removed the $gw part of the control file before the BSE calculation but forgotten that for the G0W0 calculation also "$scfinstab rpas" and "$soes all" are set, and since they appeared above the lines I added, escf ran a calculation for all possible roots with bse rather than for the lowest 3.
Title: Re: BSE excitation calculations
Post by: chris.hol on August 05, 2020, 11:42:58 AM
There is one thing to make life easier since TM 7.4 which I nearly forgot: For subsequent GW and BSE calculations in escf one can use the command line keywords -gw and -bse:
escf -gw > gw.out
escf -bse > bse.out
-gw will ignore $soes; it just defaults it to all - so one can supply the $soes group for BSE straightforward from the beginning. Further escf -gw ignores all $bse keyword groups. escf -bse then ignores all $gw and $rigw keyword groups : ). So a single control  file can be set up for a GW/BSE calculation.

All the best, Christof