[Pw_forum] Silicon HSE Problem
DHIRENDRA VAIDYA
dhirendra22121987 at gmail.com
Fri May 22 09:45:36 CEST 2015
Hi Galvin,
Checkout the units of screening parameter. I think HSE06 screening
parameter is 0.2 Angstroms^-1. And I think PWSCF accepts screening
parameter in Bohr^-1. So 0.0916 would correspond to ~0.173A^-1.
On Fri, May 22, 2015 at 11:41 AM, Khara, Galvin <galvin.khara.12 at ucl.ac.uk>
wrote:
> Hey,
>
>
> Thanks for your reply. Using an ecut of 60 Ry, a 8x8x8 kpoint grid (and
> also an 8x8x8 nqx grid), and a screening parameter of 0.0916, I got a value
> of 1.19 eV using the method you outlined. Is that a bit too good to be
> true? Heyd's original thesis quotes a value of 1.31 eV for silicon, but
> that was using a screening parameter of 0.15, which may account for the
> difference. Have you attempted to obtain bandstructure graph by using a
> defined uniform k-point list for the 2nd scf calculation? I'd like to know
> how to do this just with the eigenvalues (sorry for the novice questions),
> and thanks again for your reply.
>
>
>
> Galvin,
> University College London
> ------------------------------
> *From:* DHIRENDRA VAIDYA <dhirendra22121987 at gmail.com>
> *Sent:* 20 May 2015 20:40
>
> *To:* PWSCF Forum
> *Subject:* Re: [Pw_forum] Silicon HSE Problem
>
> Hi Galvin,
>
> I am also looking at HSE bandstructure calculaitons, mainly for Group IV
> elements from semiconductore point of view. I get from PWSCF forum that its
> difficult to obtain the bandstructure using HSE, but bandgap might be easy
> to obtain. Try this,
> In scf calculations add, nbnd (=8 if using 2 atom FCC lattice for
> Silicon). Remove 'occumpations='smearing''. And then do scf calculations
> first without HSE and then with HSE with 'restart_mode='restart''. In
> output file you will find something like this,
> highest occupied, lowest unoccupied level (ev): #some_value1
> #some_value2
>
> the difference is the bandgap = #some_value2-#some_value1
>
> On Wed, May 20, 2015 at 6:54 PM, Khara, Galvin <galvin.khara.12 at ucl.ac.uk>
> wrote:
>
>> Hey,
>>
>>
>> I'm getting a value of 5.445 Angstroms for my lattice paramter, which
>> is pretty close to the value quoted by Heyd (5.45), and that using a
>> different screening parameter (which he says does alter the physical
>> properties anyway). The next thing I want to do is calculate the band
>> gap, searching through the archives actually gives loads of topics on this
>> very problem. However, I'm really new to DFT, is there any chance you could
>> provide a method to do this, so far I've understood;
>>
>>
>> (i) - Do a regular scf calculation using an automatic fine k-point mesh
>> (I'm running an 8x8x8 simulation now)
>>
>>
>> (ii) - Run another scf calculation with a defined k-point list (does
>> this k-point list depend at all on my choice of NxNxN in (i)). Can I choose
>> any the standard 28 k-points quoted in example01?
>>
>>
>> (iii) - How do I actually calculate my band gap from the eigenvalues
>> outputted from this 2nd scf?
>>
>>
>> Sorry about my ignorance on this. To compare it with the literature
>> band gap, would it also be possible to just use the dos post processing
>> tool? I will be needing the dos at some point in the future anyway. Is this
>> used in the same way as it is for non-hybrid functionals?
>>
>>
>> Thank you so much for taking the time to read this. It's really
>> appreciated.
>>
>>
>> Galvin,
>> UCL.
>> ------------------------------
>> *From:* Layla Martin-Samos <lmartinsamos at gmail.com>
>> *Sent:* 14 May 2015 13:10
>>
>> *To:* PWSCF Forum
>> *Subject:* Re: [Pw_forum] Silicon HSE Problem
>>
>> good! check and see if it is within the litt. values.
>>
>> cheers
>>
>> Layla
>>
>> 2015-05-14 13:05 GMT+02:00 Khara, Galvin <galvin.khara.12 at ucl.ac.uk>:
>>
>>> Hey,
>>>
>>>
>>> By using a value of 0.096, (Heyd recommends using a value smaller than
>>> 0.15 to reproduce accurate band gaps in semi-conductors), my cutoff
>>> energies oscillate as follows; (scf energy, ecut energy).
>>>
>>>
>>> -14.56060124
>>> -14.56579225 -- 20 Ry
>>> -14.56576890
>>> -14.56671699 -- 30 Ry
>>> -14.56731443
>>> -14.56796231 -- 40 Ry
>>> -14.56725086
>>> -14.56720876 -- 50 Ry
>>> -14.56751976
>>> -14.56750677 -- 60 Ry
>>> -14.56749110
>>> -14.56713377 -- 70 Ry
>>> -14.56711227
>>> -14.56709019 -- 80 Ry
>>> -14.56706889
>>> -14.56704906 -- 90 Ry
>>> -14.56722966
>>> -14.56721482 -- 100 Ry
>>> -14.56700415
>>> -14.56699186 -- 110 Ry
>>> -14.56697995
>>> -14.56696887 -- 120 Ry
>>> -14.56695845
>>> -14.56694874 -- 130 Ry
>>> -14.56693973
>>> -14.56693137 -- 140 Ry
>>> -14.56692360
>>> -14.56691649 -- 150 Ry
>>> -14.56690962
>>> -14.56690322 -- 160 Ry
>>> -14.56698048
>>> -14.56697473 -- 170 Ry
>>> -14.56696928
>>> -14.56696413 -- 180 Ry
>>> -14.56693442
>>> -14.56692978 -- 190 Ry
>>> -14.56692543
>>>
>>>
>>>
>>> So the oscillation isn't massive, but I am only using a 2-atom cell. I
>>> am basically interested in calculating the energy of my system at elevated
>>> electronic temperatures (using femi-dirac smearing), I'm going to go ahead
>>> and calculate my band-gap and lattice parameters using this oscillating
>>> ecut, and see how close they are to literature. Do you think this is ill
>>> advised?
>>>
>>>
>>> Thanks,
>>>
>>>
>>> Galvin
>>> ------------------------------
>>> *From:* Layla Martin-Samos <lmartinsamos at gmail.com>
>>> *Sent:* 14 May 2015 09:57
>>>
>>> *To:* PWSCF Forum
>>> *Subject:* Re: [Pw_forum] Silicon HSE Problem
>>>
>>> Hi, I think that the safest way is to increase the dimension of your
>>> simulation cell (by replicating the unit cell along the three axes) and use
>>> the screening param that gives you the correct gap. To have uncontrolled
>>> noise may be a source of "random numbers". However, each researcher is
>>> responsible for the numbers he publish. The impact of noise depends also on
>>> the precision and accuracy that you will need or that you expect to need
>>> for further post-processing/analysis/multiscale ... . Personally, I try to
>>> have initial steps as good as I can, because for my research a
>>> not-that-good initial step has a big and negative impact in the next steps.
>>>
>>> cheers
>>>
>>> Layla
>>>
>>> 2015-05-12 15:19 GMT+02:00 Khara, Galvin <galvin.khara.12 at ucl.ac.uk>:
>>>
>>>> Hey,
>>>>
>>>>
>>>> Sorry to bring up an old topic, but from reading the HSE paper, I
>>>> need the screening parameter to be between 0.1 and 0.2 for a correct band
>>>> gap calculation in silicon. Is it a bad idea to use a value for ecut that
>>>> is oscillating (but not by a huge amount)?
>>>>
>>>>
>>>> Thanks,
>>>>
>>>>
>>>> Galvin
>>>> ------------------------------
>>>> *From:* Layla Martin-Samos <lmartinsamos at gmail.com>
>>>> *Sent:* 10 April 2015 14:39
>>>> *To:* PWSCF Forum
>>>> *Subject:* Re: [Pw_forum] Silicon HSE Problem
>>>>
>>>> Hi Galvin, maybe the screening_parameter ( HSE has two parameters
>>>> exx_fraction and screening_parameter) is not big/small (I never
>>>> remember if it is dividing or multiplying) enough for your small cell.
>>>> Maybe you could try to check the behavior of the total Energy
>>>> convergence with respect to variations of this parameter. (in
>>>> principle this parameter control the decay of the short range part of
>>>> the coulomb potential in your cell. If the short range part is not
>>>> small enough at the borders you may find funny results).
>>>>
>>>> cheers
>>>>
>>>> Layla
>>>>
>>>> 2015-04-10 15:20 UTC+02:00, Khara, Galvin <galvin.khara.12 at ucl.ac.uk>:
>>>> > ?Hey,
>>>> >
>>>> >
>>>> >
>>>> > I've been trying to use the HSE hybrid in QE to do some simulations
>>>> on Si.
>>>> > I'm finding that increasing my ecutwfc is causing the total energy to
>>>> > oscillate quite significantly. This seems to indicate a big problem
>>>> as this
>>>> > should always converge as it is variational.? When I use the plain PBE
>>>> > (Si.pbe-hgh.upf) pseudopotential without invoking any hybrid
>>>> functional the
>>>> > energy converges as expected.
>>>> >
>>>> >
>>>> > It only starts to oscillate when I add the input_dft='hse'
>>>> directive, I
>>>> > have not added anything non-standard to my input as far as I know (the
>>>> > default values for nqx, x_gamma_extrapolation, and exxdiv_treatment?
>>>> are
>>>> > used). My input file is as follows,
>>>> >
>>>> >
>>>> > "
>>>> >
>>>> > &control
>>>> > prefix='silicon',
>>>> > pseudo_dir = '~/pseudo'
>>>> > outdir='~/tmp'
>>>> > tprnfor = .true.
>>>> > tstress = .true.
>>>> > /
>>>> > &system
>>>> > ibrav = 1, celldm(1) =10.18, nat= 8, ntyp= 1,
>>>> > ecutwfc = $alat,
>>>> > occupations='smearing',
>>>> > degauss=0.002, smearing='gaussian',
>>>> > input_dft='hse'
>>>> > /
>>>> > &electrons
>>>> > scf_must_converge= .true.
>>>> > conv_thr = 1.0e-12
>>>> > mixing_beta = 0.7
>>>> > /
>>>> > ATOMIC_SPECIES
>>>> > Si 28.0855 Si.pbe-hgh.upf
>>>> >
>>>> > ATOMIC_POSITIONS crystal
>>>> > Si 0.0 0.0 0.0
>>>> > Si 0.5 0.5 0.0
>>>> > Si 0.5 0.0 0.5
>>>> > Si 0.0 0.5 0.5
>>>> > Si 0.25 0.25 0.25
>>>> > Si 0.75 0.75 0.25
>>>> > Si 0.75 0.25 0.75
>>>> > Si 0.25 0.75 0.75
>>>> >
>>>> > K_POINTS gamma
>>>> >
>>>> >
>>>> > "
>>>> >
>>>> >
>>>> > Also, I ultimately want to use the HSE hybrid to study Si at elevated
>>>> > electronic temperatures (up to 25,000 K) via fermi-dirac broadening,
>>>> is this
>>>> > unrealistic / unwise with this choice of hybrid? Any help is greatly
>>>> > appreciated.
>>>> >
>>>> >
>>>> >
>>>> > Thanks,
>>>> >
>>>> >
>>>> > Galvin K,
>>>> > Ph.D Researcher,
>>>> > University College London
>>>> >
>>>>
>>>>
>>>> _______________________________________________
>>>> Pw_forum mailing list
>>>> Pw_forum at pwscf.org
>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>>
>>>
>>>
>>> _______________________________________________
>>> Pw_forum mailing list
>>> Pw_forum at pwscf.org
>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>
>>
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://pwscf.org/mailman/listinfo/pw_forum
>>
>
>
>
> --
> --
> Dhirendra
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>
--
--
Dhirendra
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20150522/a7c3e29c/attachment.html>
More information about the users
mailing list