[gpaw-users] Re: [ase-users] Linear Response TDDFT on systems with Fermi smearing

[Michael Walter]

... just checked in how to check convergence, see

https://wiki.fysik.dtu.dk/gpaw/documentation/tddft/linear_response.html

when it's ready, or get your local copy immediately from the svn trunk.

Michael

2010/4/9 Michael Walter <Michael.Walter at fmf.uni-freiburg.de>

> Hi,
>
> 2010/4/9 Bjarke Brink Buus <bjarkebb at gmail.com>
>
>
>>
>> On 04/09/2010 12:11 AM, Michael Walter wrote:
>>
>> Hi,
>>
>> 2010/4/8 <s052810 at student.dtu.dk>
>>
>>> Hi,
>>>
>>> I have been performing LrTDDFT on MoS2 clusters and experience a
>>> significant difference in the spectrum depending on the Fermi smearing.
>>> When I perform a calculation on a specific system with Fermi temperature
>>> of 0, 0.0001 and 0.1 the result changes dramatically (the first peak
>>> moves
>>> from 1.5eV to 0.6eV with increasing Fermi smearing).
>>>
>>> I have tried to change the 'eps' value determining the minimum occupation
>>> difference for a transition to be considered, but it does not seem to
>>> make
>>> a difference. Can someone explain how LrTDDFT treats calculations with
>>> Fermi smearing?
>>>
>>
>> partial occupations also change the LrTDDFT part. The criterion to have an
>> accepted transition i->j is that the occupation number (f_i, f_j) difference
>> fulfills f_i-f_j>eps. The oscillator strength contains the pre-factor
>> |f_i-f_j|, however, giving small differences a small weight. If you just
>> look on the occurrence of peaks and not on the oscillator strength, there
>> might happen drastic changes (i.e. for partially occupied degenerate states
>> you'll get 0 eV transitions).
>>
>>
>> I am mostly interested in describing the optical band gap of the system
>> which I presume to be the first peak with clearly non-zero oscillator
>> strength.
>>
>
> the "clearly" is described by eps.
>
>
>>
>>> I do see changes in the eigenvalues for the system can it be the sole
>>> cause?
>>>
>>
>> I would expect the main source in the ground state.
>>
>>   The eigenvalues around the Fermi level are for no smearing and 0.1 eV
>> smearing,respectively:
>> band    eigenval    occ    eigenval        occ
>> 45      -6.44954        2       -6.47620        2.0000
>> 46      -6.17065        2       -6.25464        2.0000
>> 47      -5.77990        2       -5.76445        2.0000
>> 48      -5.74317        2       -5.74674        2.0000
>> 49      -5.71004        2       -5.70611        2.0000
>> 50      -4.67773        2       -4.66109        1.6920
>> 51      -4.25623        0       -4.28822        0.2332
>> 52      -4.12875        0       -4.14146        0.0590
>> 53      -3.95904        0       -3.98587        0.0128
>> 54      -3.79485        0       -3.81155        0.0022
>> 55      -3.65533        0       -3.71385        0.0008
>>
>>  I just realized I have performed the calculation with a higher eps value
>> only on the low (1e-4 eV) system and not on the 0.1 eV system which is the
>> one deviating the most, I hope this may produce similar results when it is
>> done.
>>
>
> as the width is merely a convergence helper, I would believe the width=0
> result. You'll get the transition 49->50 for width=0.1 eV, which is not
> possible for width=0.
>
>
>> On a slightly different note, I am including all occupied bands and 10
>> unoccupied bands in the linear response calculation, is that reasonable when
>> my main interest is the position of the first few peaks of the spectrum?
>>
>
> You can test this and I should write a tutorial. I may find some time for
> this task next week. You can do it by hand, by reducing jend in the
> diagonalization step:
>
> lr = LrTDDFT("lrtddft_result.dat.gz")
> lr.diagonalize(jend=57)
>
> and compare the result.
>
> Michael
>
>
>
>
> --
> ------------------------------------------
> Michael Walter
> Address: Freiburger Materialforschungszentrum
>         Stefan-Meier-Straße 21
>         D-79104 Freiburg i. Br.
>         Germany
> Tel.: +49 761 203 4758 and +49 761 203 7695
> Fax: +49 761 203 4701
> email: Michael.Walter at fmf.uni-freiburg.de
> www: http://omnibus.uni-freiburg.de/~mw767<http://omnibus.uni-freiburg.de/%7Emw767>
>
>


-- 
------------------------------------------
Michael Walter
Address: Freiburger Materialforschungszentrum
        Stefan-Meier-Straße 21
        D-79104 Freiburg i. Br.
        Germany
Tel.: +49 761 203 4758 and +49 761 203 7695
Fax: +49 761 203 4701
email: Michael.Walter at fmf.uni-freiburg.de
www: http://omnibus.uni-freiburg.de/~mw767
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20100415/062b1e67/attachment-0001.html