[CP2K-user] [CP2K:17109] Re: Density of states with k-points in CP2K
Matt Watkins
mattwatkinsuk at gmail.com
Thu Jun 9 14:18:44 UTC 2022
edit - for DOS with k-points and MPI
On Thursday, 9 June 2022 at 15:18:02 UTC+1 Matt Watkins wrote:
> Think there might be a bug fix for k-points and MPI processes in the works.
>
> On Thursday, 9 June 2022 at 14:55:48 UTC+1 nwi... at berkeley.edu wrote:
>
>> I do not get zeros for all delta_e, only when it is below a certain
>> threshold. When I take the default delta_e, it gives an energy gap of ~1.5.
>> When I increase delta_e to .01 the band gap shrinks to ~1. So there is some
>> kind of a binning problem it seems.
>>
>> I was using Fermi Dirac smearing simply to get CP2K to print out the
>> fermi energy, which it won't do in k-point calculation without it. I could
>> just read off the VBM from the list of eigenvalues, but I try to automate
>> this.
>>
>> I'm hoping to do this as a kpoint calculation so that I can get a good
>> DOS. While the energy gap is possible with gamma point only, the DOS isn't
>> really. Generally CP2K is used as a gamma-only code for large supercells,
>> but I hope to use it as a slightly more general solver since I like the
>> code and it seems to have most of the functionality needed.
>>
>> On Thu, Jun 9, 2022 at 1:28 AM Marcella Iannuzzi <marci... at gmail.com>
>> wrote:
>>
>>> For any delta_e do you get all zeros?
>>> And if delta_e = 0?
>>>
>>> About the settings, they look fine to me.
>>> Why are you using a Fermi-Dirac smearing for GaAs, it is not a metal.
>>> Anyway, this should not change anything with respect to the energy gap.
>>> Have you considered to run the same calculation with multiple cells
>>> instead of k-points?
>>>
>>> Regards
>>> Marcella
>>>
>>> On Wednesday, June 8, 2022 at 10:49:31 PM UTC+2 nwi... at berkeley.edu
>>> wrote:
>>>
>>>> Marcella, I looked further and played with delta_e and there is
>>>> something that seems incorrect to me.
>>>>
>>>> If I let delta_e -> 0 then the density column is equal to 0 everywhere.
>>>> It looks like this might be due to precision? The column only lists 3
>>>> decimal places.
>>>>
>>>> On Monday, June 6, 2022 at 1:02:40 PM UTC-7 Nicholas Winner wrote:
>>>>
>>>>> Thank you Marcella. I am clear about the Fermi level setting now.
>>>>>
>>>>> Any ideas as to why calculated gap is so large (first image)? The
>>>>> input file I have attached has pretty reasonable parameters I think:
>>>>>
>>>>> - multigrain can accommodate all basis functions
>>>>> - precision settings are above average
>>>>> - TZVP basis sets
>>>>> - Tight MP k-point grid of 16x16x16
>>>>> - EPS_SCF of 1e-6
>>>>>
>>>>> Not sure what else could be causing such a large discrepancy without
>>>>> published PBE results. Is it still an issue of interpreting the DOS?
>>>>> On Friday, June 3, 2022 at 2:38:05 AM UTC-7 Marcella Iannuzzi wrote:
>>>>>
>>>>>>
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> The Fermi energy is not well defined for electronic structures with
>>>>>> an energy gap.
>>>>>> Using the Fermi-Dirac distribution, CP2K sets the Fermi energy
>>>>>> somewhere in the middle of the gap.
>>>>>>
>>>>>>
>>>>>> The histogram calculated by activating PRINT%DOS is just counting the
>>>>>> number of eigenvalues falling into each histogram bit.
>>>>>> The partitioning of the histogram in bits is determined from the
>>>>>> DELTA_E keyword, the default of this is 0.001.
>>>>>> The final value of the density (sum of states per bit) is then
>>>>>> normalised by the number of NMO.
>>>>>> The column occupation reports the sum of the occupation numbers of
>>>>>> the eigenstates with energy falling into the corresponding bit.
>>>>>> For small DELTA_E, there are bits with zero density and zero
>>>>>> occupation number.
>>>>>> Setting DELTA_E to zero, the histogram is replaced by simply the list
>>>>>> of eigenvalues.
>>>>>>
>>>>>> When using k-point sampling the procedure is more or less the same,
>>>>>> but each contribution to the density as well as each contribution to the
>>>>>> occupation are multiplied by the weight of the corresponding k-point.
>>>>>>
>>>>>> Regards
>>>>>> Marcella
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Thursday, June 2, 2022 at 10:05:09 PM UTC+2 nwi... at berkeley.edu
>>>>>> wrote:
>>>>>>
>>>>>>> It is a static calculation, so iteration = 0. I redid the
>>>>>>> calculation with cell optimization. Didn't seem much of change. Here at the
>>>>>>> plots using the occupation first and the density second. I set fermi level
>>>>>>> to be 0, and it doesn't put it anywhere near the VBM like we expect. In the
>>>>>>> second case it looks like there is a gap, but its too big.
>>>>>>>
>>>>>>> Any knowledge maybe about the definition of fermi level? Maybe
>>>>>>> that's where I'm tripping up.
>>>>>>> [image: Screen Shot 2022-06-02 at 1.00.02 PM.png]
>>>>>>> Do you know anything about the fermi level definition?
>>>>>>> [image: Screen Shot 2022-06-02 at 12.59.51 PM.png]
>>>>>>> On Thursday, June 2, 2022 at 9:04:42 AM UTC-7 wave... at gmail.com
>>>>>>> wrote:
>>>>>>>
>>>>>>>> Hi
>>>>>>>>
>>>>>>>> It's not clear to me why you are looking at the first step. Did it
>>>>>>>> converge in just one step?
>>>>>>>>
>>>>>>>> At any rate, you'll notice a difference between &DOS and &PDOS.
>>>>>>>> PDOS returns values per orbital (s, p, d, f). If I understand it correctly,
>>>>>>>> the density is the DOS as a solid state physicist would understand it, and
>>>>>>>> occupation gives you a fractional filling at that energy. Not quite the
>>>>>>>> same thing.
>>>>>>>>
>>>>>>>> Regards
>>>>>>>> Sam
>>>>>>>>
>>>>>>>> On Wednesday, June 1, 2022 at 7:39:20 PM UTC+2 nwi... at berkeley.edu
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> Hi everyone,
>>>>>>>>>
>>>>>>>>> I know that DOS and points aren't what people first think of when
>>>>>>>>> using cp2k, but the functionality does seem to be there in some limited
>>>>>>>>> way...
>>>>>>>>>
>>>>>>>>> Anyway, I'm wondering how to properly use it, because my test case
>>>>>>>>> doesn't look right. I'm running a point calculation on GaAs with a 16x16x16
>>>>>>>>> Monkhorst-Pack pack grid and then printing the total DOS. I then take the
>>>>>>>>> columns from the DOS file (head attached) and plot energy (column 1)
>>>>>>>>> against occupation (column 3) using the reported Fermi energy from the
>>>>>>>>> output file of ~0.14 Ha. This results in a DOS where the Fermi energy does
>>>>>>>>> not line up with the VBM. What's more, if I visually inspect this DOS, the
>>>>>>>>> band gap of GaAs is >1eV. While that is closer to the experimental value,
>>>>>>>>> the reported band gap using GGA PBE is closer to 0.3eV.
>>>>>>>>>
>>>>>>>>> Questions:
>>>>>>>>> (1) What is the distinction between columns 2 and 3 in the DOS
>>>>>>>>> file? They give similar, but not equivalent looking dos plots.
>>>>>>>>> (2) Is the Fermi energy printed in the output file the same as the
>>>>>>>>> Fermi energy of the DOS, or is it defined different?
>>>>>>>>>
>>>>>>>>> I've attached my input file for reference. Thanks in advance.
>>>>>>>>>
>>>>>>>>> # DOS at iteration step i = 0
>>>>>>>>>
>>>>>>>>> Energy[a.u.] Density Occupation
>>>>>>>>>
>>>>>>>>> -0.40682067 0.0009 0.0957
>>>>>>>>>
>>>>>>>>> -0.40582067 0.0015 0.1543
>>>>>>>>>
>>>>>>>>> -0.40482067 0.0000 0.0000
>>>>>>>>>
>>>>>>>>> -0.40382067 0.0000 0.0000
>>>>>>>>>
>>>>>>>>> -0.40282067 0.0034 0.3516
>>>>>>>>>
>>>>>>>>> -0.40182067 0.0026 0.2656
>>>>>>>>>
>>>>>>>>> -0.40082067 0.0038 0.3828
>>>>>>>>>
>>>>>>>>> -0.39982067 0.0000 0.0000
>>>>>>>>>
>>>>>>>>> [image: Screen Shot 2022-06-01 at 10.32.10 AM.png]
>>>>>>>>>
>>>>>>>> --
>>> You received this message because you are subscribed to the Google
>>> Groups "cp2k" group.
>>> To unsubscribe from this group and stop receiving emails from it, send
>>> an email to cp2k+uns... at googlegroups.com.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/cp2k/b029f99c-f044-498e-a763-39f92341fec0n%40googlegroups.com
>>> <https://groups.google.com/d/msgid/cp2k/b029f99c-f044-498e-a763-39f92341fec0n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>>
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/c54a78e2-88be-4ab0-999c-36af6ffb295an%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20220609/4ecd1fb5/attachment-0001.htm>
More information about the CP2K-user
mailing list