Open Shell Systems

Deals with issues related to computation of optical spectra, solving the Bethe-Salpeter equation.

Moderators: Davide Sangalli, andrea.ferretti, myrta gruning, andrea marini, Daniele Varsano

Post Reply
User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Open Shell Systems

Post by Daniele Varsano » Fri Mar 21, 2014 8:44 am

Dear Vito,
Why is in the tutorial the acell for H2 molecule in the cube so large ( acell = 25 Bohr) ?
Because you want to simulate an isolated molecule in a supercell approximation. As yambo works in plane waves, you have periodic boundary conditions and you want to avoid spurious interaction between your cell and images. So in principle vacuum has to be added until convergence is reached. Such volume effect in GW are more problematic than in DFT. In practice coulomb cutoff technique can be used, but in order to work large supercell are neeeded.
See for instances:
Phys. Rev. Lett. 75, 818 (1995)
Phys. Rev. B 73, 205119 (2006).
Phys. Rev. B 73, 233103 (2006).
Should it not be much lower in order to reach the minimum of the total energy as a function of acell
No if you want to simulate a molecule and not a molecular crystal.
or is the value just a random number without any further meaning?
When you want to use coulomb cutoff technique there is a relation between the size of your system, cutoff range and supercell size.

Best Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

User avatar
Davide Sangalli
Posts: 610
Joined: Tue May 29, 2012 4:49 pm
Location: Via Salaria Km 29.3, CP 10, 00016, Monterotondo Stazione, Italy
Contact:

Re: Open Shell Systems

Post by Davide Sangalli » Fri Mar 21, 2014 9:44 am

Dear Vito,
1) Is Yambo capable of calculating open shell periodic systems ( open shell solids ) with doublet spin state, too ?
As Daniele tod you, yambo is capable to deal with open shell solids with doublet spin states, i.e. systems with a magnetic ground-state. The main difference against non-magnetic or closed shell systems in absorption is that the BSE matrix cannot be blocked in the two spin channels, i.e. SINGLET and TRIPLETS.

In the case of magnetic systems is the code itself which diagonalize the BSE matrix in the spin space. Remember that for magnetic systems it is not possible anymore to distinguish spin eigenstate straightforwardly for a number of reasons (i.e. spin contamination, static approximation of the kernel).
2) Does Yambo produces in the output file explicitly the singlet and triplet excitation energies for, let say, closed shell periodic systems or should one read it from the data ?
For non magnetic systems yambo solves just the SINGLET block because it is the only one which has non zero oscillator strenghts and thus which contribute to the epsilon. I see Daniele gave you a solution. Other two possible solutions:
A) compute the ground state of the system with nspin=2 (collinear spin) and then use yambo if the systems were magnetic. The BSE matrix will be diagonalized in spin space and you should get exaclty the same spectrum (TRIPLET does not contribute to the epsilon) but you will see also the triplet poles inspecting the eigen-energies.
B) run a BSE calculation with "BSENGexx= 0 eV" , which should corrispond to the case of a "c only" BSE kernel. In this case pay attention that the energies will be the triplet energies (so they are ok) while the eigen-vectors and thus the shape of the epsilon will make no sense because yambo will compute the oscillator strenght assuming you are in the SINGLET block of the BSE matrix.
3) Does BSE take care of the full spin structure if the spin-orbit interaction (coupling) is of the same order of magnitude as the electron-hole interaction ?
Yes yambo can also compute the aborption spectrum including the effect of the SO interaction. In this case you just need to start from a non-collinear ground state.
2 ) There is also a small error in the devel version of yambo :
type(levels) :: E_k_p_q

Error: Object 'e_k_p_q' at (1) must have the SAVE attribute for default initialization of a component
make[1]: *** [mod_ELPH.o] Error 1
make[1]: Leaving directory `/home/vito/my_local_copy/devel/src/modules'
make: *** [yambo_ph] Error 2
I've just checked and yambo_ph in version 3.4.1 is compiling correctly. By the way:
A) are you trying to compile yambo_ph (make yamo_ph / make all) for version 3.4.1 when you get this error? (Do you need it?)
B) try to run from the directory where the yambo source is:
svn revert -R ./
make clean_all
./configure "WITH_OPTIONS_YOU_NEED"
make yambo_ph
Do you still get the error ?
C) which compiler are you using ?

Best regards,
Davide
Davide Sangalli, PhD
CNR-ISM, Division of Ultrafast Processes in Materials (FLASHit) and MaX Centre
https://sites.google.com/view/davidesangalli
http://www.max-centre.eu/

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Open Shell Systems

Post by Daniele Varsano » Fri Mar 21, 2014 9:55 am

Dear Vito,
B) run a BSE calculation with "BSENGexx= 0 eV" , which should corrispond to the case of a "c only" BSE kernel. In this case pay attention that the energies will be the triplet energies (so they are ok) while the eigen-vectors and thus the shape of the epsilon will make no sense because yambo will compute the oscillator strenght assuming you are in the SINGLET block of the BSE matrix.
Note that the patch I sent you (for 4.0) does exactly what Davide is saying here above, the only dfference is that you can assign Triplet in BSKmod, to do that, i.e. Yambo ignore the xx part in the kernel, which is equivalento to assing BSENGexx= 0 eV.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

vitoversace
Posts: 58
Joined: Thu Mar 13, 2014 3:43 pm

Re: Open Shell Systems

Post by vitoversace » Fri Mar 21, 2014 6:49 pm

Dear Davide and Daniele many many thanks for your replies.

1 )
For non magnetic systems yambo solves just the SINGLET block because it is the only one which has non zero oscillator strenghts and thus which contribute to the epsilon
I did bse run in 3.3.0 version on LiF which is a diamagnetic material : For singlet

Code: Select all

 BSresKmod= "xc"
     BScplKmod="xc" 
and also for triplett

Code: Select all

BSresKmod= "c"
     BScplKmod="c"
I get for both from structure point of view very well shaped and meaningful spectra but the only difference between them is that they are a bit shifted. (Triplet is shifted toward lower energies as one would expect it.)

2)
run a BSE calculation with "BSENGexx= 0 eV" , which should corrispond to the case of a "c only" BSE kernel. In this case pay attention that the energies will be the triplet energies (so they are ok) while the eigen-vectors and thus the shape of the epsilon will make no sense because yambo will compute the oscillator strength assuming you are in the SINGLET block of the BSE matrix.
So you mean that

Code: Select all

BSresKmod= "c"
    BScplKmod="c"
means calculation of singlet spectrum but the eigen-energies that i get using ypp are triplett excitation energies ????? (very strange!!!!)

3) in order to make it clear, i am talking about this BSE hamiltionian matrix : http://rohlfing.physik.uni-osnabrueck.d ... n_2000.pdf on page 33 and singlet-triplet spin states between electron and hole of a non magnetic system like LiF.
4) in the same Habilitation on page 58 in the table you see the first TRIPLET excitation state at 12.4 eV and the first SINGLET excitation state at 12.8 eV. So i strongly assume that there is a one absorption optical spectrum for SINGLET and one for Triplet. If so, i don't get the point why in a non magnetic system only singlet states contribute to epsilon !!!!!


5)
Note that the patch I sent you (for 4.0) does exactly what Davide is saying here above, the only dfference is that you can assign Triplet in BSKmod, to do that, i.e. Yambo ignore the xx part in the kernel, which is equivalento to assing BSENGexx= 0 eV.
Very good. I did this with 3.3.0 and i got a triplet absorption spectrum which was shifted in comparison to singlet one but Davide is saying that triplet states do not contribute to optical spectrum in non magnetic materials.



So once we cleared this, i will come to other points.
With best wishes
Vito Versace
PhD student
University of York
vitoversace@yahoo.com

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Open Shell Systems

Post by Daniele Varsano » Fri Mar 21, 2014 10:10 pm

Dear Vito,
1) OK, it is right for 3.3
2)

Code: Select all

BSresKmod= "c"
    BScplKmod="c"
is 3.3
or equivalently

Code: Select all

"BSENGexx= 0 eV"
in 3.4. Yambo does the same calculation, it is just that variable changed in 3.4.
This is a calculation to get triplet excitation, it is not a singlet spectrum as you are dropping the 2Kx term in the Hamiltonian (see PRB 62 4927 (2000) or the thesis you posted the link), so they are singlet-triplet transitions. While the spectrum for the singlet is meaningful, only the excitation energies make sense for triplet as singlet-triplet transition are dipole forbidden (selection rule Delta S=0).
4) As before: Singlet - triplet transitions are not dipole allowed as they requires a forbidden spin transition. So you do not have an absorption optical spectrum for Triplet (you could have tirplet-triplet, but ground state is a singlet). This also explains why they do not enter in the epsilon.
5) You will have shifted energy level if any (singlet-triplet splitting), usually small for crystal and big for molecules, but singlet-triplet transitions do not contribute to optical spectra, indeed Michael Rohlfing reports energy level and not "absorption spectra" for triplets.

Hope it helps,

Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

vitoversace
Posts: 58
Joined: Thu Mar 13, 2014 3:43 pm

Re: Open Shell Systems

Post by vitoversace » Fri Mar 21, 2014 10:23 pm

Dear Daniele,
Okay very good.
1 ) But let me please know what is exactly the spectrum for

Code: Select all

BSresKmod= "c"
BScplKmod="c"
in 3.3.0. Is this spectrum coming from triplet-triplet transitions having a singlet ground state ? Or what is this spectrum which is a bit shifted toward lower energies when exchange is zero ???

2)
indeed Michael Rohlfing reports energy level and not "absorption spectra" for triplets.
and these triplet energy levels for instance for the one in the thesis i get when performing a full diagonalization of bse matrix ??? Right?

Bests
Vito Versace
PhD student
University of York
vitoversace@yahoo.com

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Open Shell Systems

Post by Daniele Varsano » Fri Mar 21, 2014 10:46 pm

Dear Vito,
NO, it is not. Such spectrum should not be considered.
As said by Davide only the excitation energies are meaningful as the oscillator strengths are meaningless see the quoted text :
B) run a BSE calculation with "BSENGexx= 0 eV" , which should corrispond to the case of a "c only" BSE kernel. In this case pay attention that the energies will be the triplet energies (so they are ok) while the eigen-vectors and thus the shape of the epsilon will make no sense because yambo will compute the oscillator strenght assuming you are in the SINGLET block of the BSE matrix.
We should remove the output of the spectra when a triplet calculation is performed, as it can create confusion.
To resume the Hamiltonian using:

Code: Select all

BSresKmod= "c"
BScplKmod="c"
is the triplet hamiltonian, the diagonalization gives you triplet excitation energies, so energy differences between ground state (singlet) and the triplet states, but the spectrum itself is meaningless as the oscillator strengths are fake. You can look at the excitation energies by using "ypp -e s"
after diagonalization and you should find the shifts as reported by Rohlfing.

Beside that, note that in many systems it is useful to look at the excitation energies also for the singlet (beside the spectrum) as may be you can be interested in dark excitons or dark states, ie excited states with zero oscillator strength that you cannot see in the spectrum.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

vitoversace
Posts: 58
Joined: Thu Mar 13, 2014 3:43 pm

Re: Open Shell Systems

Post by vitoversace » Sat Mar 22, 2014 2:07 am

Dear Daniele,

I already got the point... thanks a lot for clarifying things.
Yes the triplet output was indeed very very confusing for me over the last days.

1) You mentioned :
(you could have triplet-triplet, but ground state is a singlet
I assume that currently at this stage of code development this is not implemented to be calculated ? Right?

2)
In the case of magnetic systems is the code itself which diagonalize the BSE matrix in the spin space. Remember that for magnetic systems it is not possible anymore to distinguish spin eigenstate straightforwardly for a number of reasons (i.e. spin contamination, static approximation of the kernel).
So this means that for a magnetic system one can not refer the obtaind eigenstates of bse hamiltonian to any ordinary pure spin states like doublet, quartet, etc ... ? Right ?
So the eigenstates of bse hamiltonian for magnetic materials are just mixed states of spin states with multiplicity 2,4, and so on ... Right?

3) Just for being absolutely sure and avoiding any misunderstandings : If the spin orbit in a NON-magnetic system is strong, the code diagonalizes bse matrix with the FULL SPIN STRUCTURE ? (NO decoupling any more
between singlet and triplet channel) Right?

4) Since only the static part of screening enters to bse calculation, how much would be the effect of the absence of dynamic screening roughly ?
(so for instance the absence of local field effects in calculations means 10 to 15 percent increase in intensities of spectrum roughly.
So what about the dynamic screening or is it very system depending and no estimation can be done on this ? )

Bests
Vito Versace
PhD student
University of York
vitoversace@yahoo.com

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Open Shell Systems

Post by Daniele Varsano » Sat Mar 22, 2014 8:54 am

Dear Vito,

1) it was just an example of optical allowed excitation. You can have the triplet energies, why do you want to calculate the strengths? Do you know experiments that measure that?

2) Yes, I do not know if high spin configuration can be recognized, may be Davide can gives you details.

3) Yes

4) Very system dependent. As also the inclusion of local fields effects in my opinion.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

vitoversace
Posts: 58
Joined: Thu Mar 13, 2014 3:43 pm

Re: Open Shell Systems

Post by vitoversace » Mon Mar 24, 2014 1:02 am

Dear Daniele,

1) because i am interested in theoretical spectroscopy of solids based on mbpt. As much as i know these are laser and/or Fourier spectroscopy but how to measure it is none of my business.
2) maybe a question for Davide: If the eigenstates of bse matrix of a magnetic material are mix of different spin states, so the excitation energies are practically wrong because of spin contamination or not ?
3) the answer to high spin configurations is of great importance to me.
4) is there any thumb rule to guide one when the inclusion of coupling term in bse matrix is important and when not or again it is a case of testing?

Bests
Vito Versace
PhD student
University of York
vitoversace@yahoo.com

Post Reply