BSE calculation

Message

giacomo giorgi · #1 Post by **giacomo giorgi** » Sat Feb 16, 2013 7:12 am

Dear Vasp Admin,
I am facing a problem with vasp5.3.2 in BSE calculation.
Every time I try to make my calculation at BSE level on top of the previous and correctly terminated GW one I get that my job ends without any error message.

I am both using 18 and 36 processors, thus I suppose it is not a problem of memory.
I used 6x6x1 k-point sampling and 4 bands of valence and 6 in the conduction, thus the BSE matrix is supposed to be not so huge.

this is the point where job ends

1) OUTCAR:
...

total amount of memory used by VASP on root node 426039. kBytes
========================================================================

base : 30000. kBytes
nonlr-proj: 7871. kBytes
fftplans : 1577. kBytes
grid : 15015. kBytes
one-center: 373. kBytes
HF : 259. kBytes
wavefun : 0. kBytes
response : 347341. kBytes
bse : 23603. kBytes

BSEDIAG: cpu time 22.34: real time 22.37
BSEOSZI: cpu time 1.11: real time 1.12

2) vasp.log

...

BSE single prec attempting allocation of 0.024 Gbyte rank= 1718
BSE setting up matrix
reading now WFULL0001.tmp
reading now WFULL0019.tmp
|. reading now WFULL0003.tmp
. reading now WFULL0015.tmp
.......|.........|.........|......... BSE redistributing all elements

BSE diagonalizing matrix (ZHEEVX)
BSE calculating oscillator strength

and this is my INCAR

SYSTEM = Si_Vac_Td

# xc
GGA = PE
VOSKOWN = 1
ISTART = 1
#electronic relaxation
LREAL = Auto
PREC= High

ISPIN=2
MAGMOM=4*3.0 8*0.

IBRION = 1

#accuracy of the calculation
EDIFF = 1.E-5
EDIFFG = -0.005 ! inizialmente -0.004

ENCUTGW=100
LOPTICS = .TRUE.
NBANDS = 252

ISMEAR=0
PRECFOCK=Normal

ALGO = BSE
NOMEGA = 50
NBANDSO = 4 ! number of bands for electron-hole treatment (occupied)
NBANDSV = 6 ! number of bands for electron-hole treatment (virtual)
CSHIFT=0.05
OMEGAMAX=6 ! max frequency
NEDOS=2000 !number of energy points

NGX=44
NGY=44
NGZ=66

#ionic steps
NSW = 0

AMIX = 0.2
BMIX = 0.00001
AMIX_MAG = 0.8
BMIX_MAG = 0.00001

LORBIT = 11

I would be extremely grateful if you could help me in fixing this problem

Thanks in advance,
Giacomo

giacomo giorgi · #2 Post by **giacomo giorgi** » Thu Feb 21, 2013 1:32 am

I reduced the NBANDS=72 and NBANDSO=NBANDSV=1 but same result.... please help me!

#3 Post by **admin** » Thu Feb 21, 2013 5:25 pm

LOPTICS=.TRUE. is in BSE calculations not used.
NOMEGA value should be dividable by the number of computer nodes.

huangj3 · #4 Post by **huangj3** » Thu Feb 21, 2013 7:45 pm

Dear admin,

By "the number of computer nodes", I suppose it means the number of cores. If a node has 24 cores, and two nodes are used, then NOMEGA should be set to 48 or 96 (to be within the 50-100 range suggested on the manual). Is that correct?

Thanks.

huangj3
[ Edited Thu Feb 21 2013, 07:47PM ]

giacomo giorgi · #5 Post by **giacomo giorgi** » Fri Feb 22, 2013 3:45 am

Dear Admin,
thanks. But regretfully I got the same behaviour without LOPTICS=.TRUE. and NOMEGA=72 (proc number 18, NBANDS=288; NBANDSO = 1; NBANDSV = 1)

Any other idea?

Thanks

giacomo giorgi · #6 Post by **giacomo giorgi** » Fri Feb 22, 2013 5:50 am

Dear Admin,
I reduced to the very minimal setup my INCAR.... no way.....still frozen job in the same point....
Sorry and thanks

SYSTEM = Si_Vac_Td

# xc
GGA = PE
VOSKOWN = 1
ISTART = 1
LREAL = Auto
PREC= High

SIGMA = 0.05
ISPIN=2
MAGMOM=4*3.0 8*0.

IBRION = 1
EDIFF = 1.E-5
EDIFFG = -0.005 ! inizialmente -0.004

NBANDS = 72

NOMEGA = 48
ENCUTGW=100
SIGMA = 0.05 !this is small smearing of 0.1eV
PRECFOCK= N

ALGO = BSE
NBANDSO = 1 ! number of bands for electron-hole treatment (occupied)
NBANDSV = 1 ! number of bands for electron-hole treatment (virtual)
CSHIFT=0.05
OMEGAMAX=6 ! max frequency
NEDOS=2000 !number of energy points

NGX=44
NGY=44
NGZ=66

#ionic steps
NSW = 0

AMIX = 0.2
BMIX = 0.00001
AMIX_MAG = 0.8
BMIX_MAG = 0.00001

LORBIT = 11

yunhailiseu · #7 Post by **yunhailiseu** » Sun Mar 24, 2013 2:19 pm

I am troubled this problem, too.

As far as I know, when vasp reports BSE 'calculating oscillator strength', the whole calculation is almost over. And if no error message is given after that, the job just finished sucessfully ( Line841 - Line870 of bse.F ).

However, the output is not written in OUTCAR as in the usual case. The oscillator strength and dielectric tensor is written in vasprun.xml, and post-process may be performed in order to obtain absorption spectra.

May this be of help.
[ Edited Sun Mar 24 2013, 04:10PM ]

giacomo giorgi · #8 Post by **giacomo giorgi** » Mon Apr 01, 2013 3:59 am

Dear yunhailiseu,
thanks a lot. It is relaxing that it is not only my problem. But let me ask you, how to postprocess in order to get absorption spectra?
Thanks once more,
Giacomo

yunhailiseu · #9 Post by **yunhailiseu** » Tue Apr 02, 2013 1:25 pm

Absorption spectra is determined by the imaginary part of the dielectric function.

How to obtain the dielectric function along a specific direction depends on the symmetry of the crystal.

Take the hexagonal lattice for example. The direction of the electric field of the incident light may be parallel or perpendicular to c axis of the crystal, and the dielectric function in these two cases are:
eps = eps_zz and eps = 1/2 * ( eps_xx + eps_yy ).
where eps_xx, eps_yy and eps_zz are components of the dielectric tensor.

You may find these formulas in published papers or textbooks on solid state physics or nonlinear optics.

giacomo giorgi · #10 Post by **giacomo giorgi** » Tue Apr 16, 2013 9:07 am

Dear yunhailiseu,
thanks a lot!
But then let me ask you...how to understand if an exciton is bound or not? I want to say no possibility of having a BSE calculated bandgap..... Am I correct?
thanks a lot!
Giacomo

yunhailiseu · #11 Post by **yunhailiseu** » Wed Apr 17, 2013 8:40 am

Usually the BSE calculation is performed on top of GW calculation. From the output of GW calculation the exact bandgap Eg can be determined.

Then in the following BSE calculation the excitation energies can be determined ( in vasprun.xml just before the dielectric tensor, together with transition matrix elements ). If one exciton has a corresponding excitation energy lower than Eg, it is a bound exciton. And if the excitation energy is larger than Eg, it is a resonant exciton.
The type of exciton may also be determined from absorption spectrum.

You may consult the following papers : Nano Lett. 2010, 10, 426-431, Nano Lett. 2007, 7, 3112-3115 and PRB 83, 085405 (2011).

[ Edited Wed Apr 17 2013, 08:41AM ]

giacomo giorgi · #12 Post by **giacomo giorgi** » Thu Apr 25, 2013 9:26 am

Dear yunhailiseu,
Thanks for the very clear and prompt reply. Thanks for the two references. I read them. I have anyway a couple of doubts that remain unsolved.
1) as far as I know, there is an optical band gap you calculate from bse calculation and that you can compare to that obtained experimentally by adsorption experiments (where e-hole interactions are considered). The experimental counterpart of the theoretically calculated GW electronic bandgap consists at variance in PL emission experiment where the electron is ejected (direct mechanism) and the electron-hole interactions are neglected. Thus, I suppose that also a BSE calculated optical bandgap must be deduced from Outcar.
For example, In the case of layered system a dft/rpa spectrum is calculated. A gw correction is applied that redshifts the spectrum. Including local fields and e-h Interactions (bse) a new spectrum is calculated which is blue shifted with respect to the Gw one, due to quantum confinement. You can easily check it by comparing the imaginary part of the dielectric function calculated at the three levels of theory.

2) in calculating the bse spectrum with vasp, how to treat magnetic systems? My VB is spin up and the CB is spin down. Does the optical transition matrix take into account this? Or is it the up-up transition taken as bandgap?

I am very interested in your opinion about these two points.
Thanks a lot and excuse me the long post!!!

Best,
Giacomo

yunhailiseu · #13 Post by **yunhailiseu** » Fri Apr 26, 2013 6:12 am

There are two basic electronic excitations : charged excitation in which one electron is removed from or added to the system, and neutral excitation in which the number of electrons remain unchanged. GW method is designed to deal with the former, while BSE for the later.

Supposing a N-electron system, in principle there exists its ground-state which can be as|N,0> and excited states |N,i> (although solving Schrodinger's equation for N electrons is impractical). Analogously, for N-1 & N+1 systems we have |N-1,i> & |N+1,i>.

The charged excitation can be described as |N,0> -> |N-1,i> ( as in photoemission experiments ) and |N,0> -> |N+1,i> ( as in inverse photoemission experiments ). The energy difference between |N,0> and |N+1,i> defines the excitation energy, which in theory can be obtained by solving Dyson's equation, i.e. the GW method.

Neutral excitation can be decribes as |N,0> -> |N,i>( as in photo-absorption ), and excitation energy can be determined by solving Bethe-Salpeter Equation. The inverse process of neutral excitation, i.e. Photoluminescence can be denoted as |N,i> -> |N,0>.

Since GW method and PL correspond to different excitations respectively, differences between results obtained from the two are likely to occur.

The term 'optical gap' may be improper, since the wavefunctions of e-h pair are not k-point dependent and do not form bands. The actual neutral excitation spectrum is more like energy spectrum of atom or molecule orbits. So "lowest excitation energy" may be better.

Usually GW method enlarges Eg from dft results, while e-h interaction reduces lowest excitation energy from Eg. So GW-RPA spectra is often blue shifted from DFT-RPA, not red shifted. And BSE spectra is red shifted from GW-RPA spectra, not blue shifted. Moreover, the shape of spectra of GW and dft may be much alike, while BSE spectra may differ greatly from the two due to mix of transitions.

I don't know how to treat spin-polarized system in vasp either .Sorry for that.

[ Edited Fri Apr 26 2013, 06:14AM ]

giacomo giorgi · #14 Post by **giacomo giorgi** » Fri Apr 26, 2013 8:50 am

Thanks once more.

Sorry, I always confuse red and blue shift... of course you are right!
Thanks!
Giacomo

Isabel · #15 Post by **Isabel** » Fri Jun 28, 2013 9:14 pm

Hi Giacomo,

I'm facing some problems to perform BSE calculations with VASP code. Can you please give me your email address then I can contact to directly?
Many thanks in advance.

Isabel