Yambo vs Gaussian

Run-time issues concerning Yambo that are not covered in the above forums.

Moderators: Daniele Varsano, andrea marini, Conor Hogan, myrta gruning

Yambo vs Gaussian

Postby martinspenke » Tue May 20, 2014 11:20 pm

Dear Daniele,

Following problem :

I tried to cross-check yambo results for an isolated geometry optimized furan molecule (consisting of only 9 atoms) in a 20 bohr large box vs gaussian results ( coupled cluster method ).
Unfortunately yambo produces quite very wrong excitation energies although i have quite converged parameter values !

I really have no idea why this is the case. A bit disappointing.

Best Regards
Martin
You do not have the required permissions to view the files attached to this post.
Martin Spenke, PhD Student
Theoretisch-Physikalisches Institut
Universität Hamburg, Germany
martinspenke
 
Posts: 149
Joined: Tue Apr 08, 2014 6:05 am

Re: Yambo vs Gaussian

Postby Daniele Varsano » Wed May 21, 2014 4:17 am

Dear Martin,
Very quickly, next I will look again in details at your calculations.
1) For finite systems it is needed to insert the coupling part of the Hamiltonian (BSEmod=coupling) and uncomment #WehCpl, check the literature on that (eg this paper), Tamm-Dancoff approximation most probably it is not valid.
2) In finite systems, being in plane waves, so periodic boundary conditions, you need to cut the coulomb potential in order to avoid
spurious interactions between repeated cells. Not doing that can have important biases, in particular in the GW gap (there is a lot of literature on that, for instance have a look here). You can activate it in with yambo -r (I suggest you to use a sphere for this case).
3) You are using a real axis integration (very hard to converge!!) try also a Plasmon Pole approximations, even if not well justified, very often it gives the right answers and it is easier to converge than the real axis. Just to have a look on how much they differs, anyway have a look in the literature if you can have some information on the QP gap (both theoretical and experimental), just to individuate the poor calculation if GW or BSE (or may be both).
4) Convergences, check and recheck (eg BSENGBlk it looks me quite low with respect the total G vectors, but it is just an impression, not motivated).
5) BSE and CC are very different level of theory, so it is not guarantee they should give the same results, for instance if double excitation character excited states are present. Anyway I agree with you that the in this case you found a very big discrepancy.

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
Daniele Varsano
 
Posts: 1975
Joined: Tue Mar 17, 2009 2:23 pm

Re: Yambo vs Gaussian

Postby martinspenke » Wed May 21, 2014 10:01 am

Dear Daniele,

Thanks a lot.
I will check your points in a day or so, and will let you know.
Hopefully this big discrepancy is not due to double excitations and hence the necessity of using dynamical kernel.

Bests
Martin
Martin Spenke, PhD Student
Theoretisch-Physikalisches Institut
Universität Hamburg, Germany
martinspenke
 
Posts: 149
Joined: Tue Apr 08, 2014 6:05 am

Re: Yambo vs Gaussian

Postby martinspenke » Sat May 24, 2014 2:30 pm

Dear Daniele,

I checked your points, but it did not help.
I think the source of the problem is that in yambo one calculates an isolated molecule
in a box WITH screening. This big discrepancy in energies can only be explained by screening effects in the box which blue shift the spectrum
besides coulomb interactions which i think, should be relatively small in a big box or using r_cut.

I also tested it for an even more simpler molecule, namely for water in a 30 bohr box,
and cross-checked the results vs SAC-CI calculations, and got very wrong results produced by yambo.

Now the question is whether there is a possibility to turn off screening calculation in yambo ?

Bests
Martin
Martin Spenke, PhD Student
Theoretisch-Physikalisches Institut
Universität Hamburg, Germany
martinspenke
 
Posts: 149
Joined: Tue Apr 08, 2014 6:05 am

Re: Yambo vs Gaussian

Postby Daniele Varsano » Sat May 24, 2014 4:11 pm

Dear Martin,
I can't see your point, screening it is an ingredient of the GW and BSE equation, and it is needed in the Hedin Equations.
In any case, if you want to remove the screening you can do it: please note: Bethe-Salpeter equation without screening it is totally equivalent to the Time Dependent Hartree Fock, so you need to build up your input by using (yambo -k hf).
I do not know what kind of agreement you are aiming at, but bear in mind that you are comparing:
multideterminant vs singel determinant methods
plane wave vs gaussian basis set
all electron vs pseudopotential methods
So there could be many sources for the discrepancies.
There are several examples of BSE with yambo giving nice agreement with experiments for molecules, you can have a look for instance to these papers:
http://nano-bio.ehu.es/ab-initio-electr ... orrelation
http://www.yambo-code.org/papers/nl803717g.pdf
http://link.aip.org/link/?JCP/138/024312
http://pubs.acs.org/doi/abs/10.1021/ct3007502
http://www.sciencedirect.com/science/ar ... 1X14001261
You can have a look at the convergence parameters, that for isolated molecules need to be pushed a lot: in some cases they are in the supporting info.

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/
User avatar
Daniele Varsano
 
Posts: 1975
Joined: Tue Mar 17, 2009 2:23 pm

Re: Yambo vs Gaussian

Postby martinspenke » Sun May 25, 2014 11:52 pm

Dear Daniele,
many many thanks for these valuable papers, i red most of them. It was very useful.

1) So the point is that i would like to produce some absorption peaks of a single water molecule in a cubic box.
Using yambo i can nearly confirm the results of these people in this paper http://journals.aps.org/prl/abstract/10.1103/PhysRevLett.100.207403
(they applied SAC-CI) for a single water molecule which is screened in order to simulate environmental effects in solid ice (Fig.2 (b) in the paper, page 3).
(I get peaks at 8.5, 10.5, 12.9, 14.2 using YAMBO, so every thing okay so far.)

But unfortunately i can not generate the 7.2 eV peak for a single water molecule in a box WITHOUT SCREENING using Yambo (Fig.2 (a), page 3).
This screening of a water molecule should simulate the optical absorption spectrum of ICE.
Is there now any possibility to get the spectrum of a single water or its excitation energies without screening applying BSE in order to be able to compare the energies with usual multi determinant quantum chemistry methods or not???

Or let my ask it in this way : how can i calculate excitation energies of a single water molecule in pure vacuum applying BSE in YAMBO ?????


2) Another problem is, that when i use the coulomb cutt-off technique with a radius of 20 bohr using a sphere in gw calculation for a water molecule in the box,
i get very very wrong quasi particle energies for occupied states !!!


Best Regards
Martin
You do not have the required permissions to view the files attached to this post.
Martin Spenke, PhD Student
Theoretisch-Physikalisches Institut
Universität Hamburg, Germany
martinspenke
 
Posts: 149
Joined: Tue Apr 08, 2014 6:05 am

Re: Yambo vs Gaussian

Postby Daniele Varsano » Mon May 26, 2014 8:56 am

Dear Martin,
First of all, water molecule it is a well known hard task, I do not know if you are interested in water or you are using it as example, in the latter case you choose an hard example.
If you are interested in water in BSE you can have a look to this paper:
PRL 97, 137402 (2006)
About screening: as I told you before, screening it is an essential ingredient of BSE, if you discard it you are solving the TD-HF equation, I do not know if this is what you want to do. Of course screening effects are different in an isolate case or in a crystalline phase. If you want simulate ice , you should know anyway how your unit cell it is done, and lattice parameters. Macroscopic screening, in an isolated case goes to 1 (indeed it is what happen when you introduce a coulomb cutoff), but still you have non zero component of the screening matrix (local screening).
In your qp calculation, as you say something went totally wrong in the exchange part of the self-energy, I do not know what is the problem, but if you set a 20au cutoff, your supercell side has to be at least 40au (do not forget you are in plane wave!!! and if you do not do that, you are considering a cutoff larger than your periodicity that is a nonsense). You can have a look to this paper.
Other two comments/suggestions:
1) 500 Gvectors in the dielectric matrix looks me a bit few, for a supercell calculation of isolated molecules.
2) Real-axis in my experience is tremendously hard to converge, why don't you give a try with the plasmon-pole model?

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/
User avatar
Daniele Varsano
 
Posts: 1975
Joined: Tue Mar 17, 2009 2:23 pm

Re: Yambo vs Gaussian

Postby martinspenke » Mon May 26, 2014 12:10 pm

Dear Daniele,

The point is, i am just wondering how people in papers presenting excitation energies for instance here : http://ftp.aip.org/epaps/journ_chem_phys/E-JCPSA6-139-020340/SI.PDF , table 3 on page 39
for isolated water molecule using BSE which as you said contains screening effects and compare these results with those of ordinary quantum chemistry methods which do not have any screening ingredients
and the results of both methods are quit in very good agreement ??? !!!???

In table 3 on page 39 of the above paper, as you see, they get this 7.2 eV peak using TDA-BSE for water molecule. How can one get this, too, using yambo for instance ???

1) Yes screening effects for isolated water molecule and ice are different, but it seems that you can nearly simulate the peak positions of ice spectrum with one single water molecule, too.
(I calculated it for a hexagonal Ice Ih, too. The peak positions of both systems are nearly the same. Yes the shape of both spectrum is different, but still you see some similarities, as the first peak is in both cases quit intense.)

2) I tested some other Gvectors values, but no significant change any more in qp energies.
3) I tested PPA, too, i nearly get the same qp energies, and i can get quit converged qp using real axis integration,too. So this is not my problem, at least for an isolated water molecule.

SO AGAIN MY CORE QUESTION as i am interested in water molecule :
How can i get the 7.53 eV singlet peak of water molecule in gas phase using BSE in YAMBO as other people in the above paper have done this already ?


With best wishes
Martin
Martin Spenke, PhD Student
Theoretisch-Physikalisches Institut
Universität Hamburg, Germany
martinspenke
 
Posts: 149
Joined: Tue Apr 08, 2014 6:05 am

Re: Yambo vs Gaussian

Postby Daniele Varsano » Mon May 26, 2014 12:37 pm

Dear Martin,

The point is, i am just wondering how people in papers presenting excitation energies for instance here : http://ftp.aip.org/epaps/journ_chem_phy ... 340/SI.PDF , table 3 on page 39
for isolated water molecule using BSE which as you said contains screening effects and compare these results with those of ordinary quantum chemistry methods which do not have any screening ingredients
and the results of both methods are quit in very good agreement ??? !!!???


Two different ways to account electronic correlation: Green function based: diagramatic expansion with respect W (screened coulomb potential), wave function based, summing up determinats.

but it seems that you can nearly simulate the peak positions of ice spectrum with one single water molecule, too.

OK, remember that it is not a single molecules, in plane waves you have always a crystal, it would be single molecule for infinite supercell volume, ot trying to account for it via a coulomb cutoff

Molecule in gas phase in yambo:
1)Put the molecule in a big box.
2)Calculate GW corrections: checking carefully convergences (Gvec and bands both in G and in X), volume effects are important!!! the coulomb cutoff it helps (do not know what is going wrong now with your calculation). If you do not want to use the cutoff, use at least the random integration (RandGvec=1, RandQpts=1000000), but verify you are near convergences considering different volumes. Or you can also try to use a box cutoff, in this case the Random integration is mandatory, and consider that for internal reason in the code, in input it is needed to set the size of the box double with respect the one you want to consider (this is something I need to fix).
3) If there are previous calculations in the literature compare your GW gap, if it is not in agreement your excitations energies will be shifted.
4) Do a BSE calculation, if you want a TD approximation use BSEmod="resonant", in this case the calculation will be much faster and sometimes TDA approx gives better results than the full BSE.

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/
User avatar
Daniele Varsano
 
Posts: 1975
Joined: Tue Mar 17, 2009 2:23 pm

Re: Yambo vs Gaussian

Postby martinspenke » Mon May 26, 2014 5:04 pm

Dear Daniele,

I have done all these steps with ( cut-off works now for occupied states, too.) or without cut-off using a box size of 25 au many times, pushing high all parameters (Gvectors, bands, etc,...)
but i never ever get this first peak at 7.4 eV (experiment) or at least near to it. I get for the first peak 8.6 eV which matches the case of "isolated water molecule" with local screening.
How can i get this 7.4 eV peak in the case of isolated water molecule without local screening using BSE ???

And also for other molecules, naphthalene, furan, etc excitation energies are all blue shifted due to local screening and i can not compare it with available qc methods !!!

I am sorry if i am annoying you but i would like very much to compare mbpt vs other qc methods.

Bests
Martin
Martin Spenke, PhD Student
Theoretisch-Physikalisches Institut
Universität Hamburg, Germany
martinspenke
 
Posts: 149
Joined: Tue Apr 08, 2014 6:05 am

Next

Return to Other issues

Who is online

Users browsing this forum: No registered users and 1 guest