[QE-users] Potential issue with charge density cutoff convergence and K-point grid convergence

Giuseppe Mattioli giuseppe.mattioli at ism.cnr.it
Thu Apr 26 09:04:55 CEST 2018


Dear Martina
If you look at Jia’s comment from the other side, a norm conserving  
pseudo might be OK with a 80/320 Ry cutoff (ONCV PPs from the sg15  
library usually are). And you get rid of all the usual limitations of  
US (e.g. slow or unstable calculations with hybrid EXX functionals).
HTH
Giuseppe

Martina Lessio <ml4132 at columbia.edu> ha scritto:

> Dear Jia,
>
> Thanks so much for your feedback. I see your point, although that is the
> result I got from my convergence tests. Perhaps I will have another look at
> my tests and check that I have not been too conservative with selecting the
> parameters. Furthermore the suggested minimum cutoff for one of the pseudo
> potentials I am using is 50 Ry so I don't think I can go much lower than
> that.
>
> All the best,
> Martina
>
> On Wed, Apr 25, 2018 at 5:23 PM, Jia Chen <jiachenchem at gmail.com> wrote:
>
>> Dear Martina,
>>
>> If I may interject, the advantage of ultrasoft pseudopotentail is not
>> using wavefunction cutoff as large as 55 Ry. Usually ultrasoft and norm
>> conserving pseudopotential have similar density cutoff, but ultrasoft only
>> need about half the wavefunction cutoff.
>>
>> Cheers
>> Jia
>>
>> On Wed, Apr 25, 2018 at 4:11 PM, Martina Lessio <ml4132 at columbia.edu>
>> wrote:
>>
>>> Dear Stefano,
>>>
>>> I have one more question regarding the convergence tests. I have
>>> performed the tests in the order that you recommended and found that if I
>>> set ecutrho=280 Ry I can get away with ecutwfc=55 Ry without negatively
>>> impacting the convergence. So I am thinking of using these parameters for
>>> my future calculations. However, I have read in many QE resources that if
>>> you use ultrasoft pseudopotentials like I do, ecutrho should be at least
>>> equal to 8*ecutwfc. So I am now wondering whether it is an issue that in my
>>> case ecutrho is only equal to about 5*ecutwfc or given that I performed the
>>> recommended convergence tests it is okay to have this setup even for
>>> pseudopotentials.
>>>
>>> I hope my question is clear.
>>> Thanks so much!
>>>
>>> All the best,
>>> Martina
>>>
>>> Martina Lessio
>>> Postdoctoral Research Scientist
>>> Department of Chemistry
>>> Columbia University
>>>
>>>
>>> On Tue, Apr 24, 2018 at 10:53 AM, Martina Lessio <ml4132 at columbia.edu>
>>> wrote:
>>>
>>>> Dear Stefano,
>>>>
>>>> Thank you very much for your prompt response, that helps a lot!
>>>> I will repeat the tests as you suggested but it's good to know that my
>>>> system is behaving normally and I can then proceed with more complex
>>>> calculations.
>>>>
>>>> All the best,
>>>> Martina
>>>>
>>>> On Tue, Apr 24, 2018 at 3:23 AM, Stefano de Gironcoli <degironc at sissa.it
>>>> > wrote:
>>>>
>>>>> Dear Martina Lessio,
>>>>>
>>>>>    first of all I would say that a convergence error of 1.d-5 Ry in a 6
>>>>> atom cell looks pretty much converged to me. I think that even  
>>>>> your ecutrho
>>>>> = 240 Ry calculation (~1.d-5 Ry per atom) looks quite good.
>>>>>
>>>>>    coming to the way things converge:
>>>>>
>>>>>    - total energy convergence with respect to ecutwfc is expected to be
>>>>> from above because of variational principle: the higher ecutwfc is the
>>>>> more planewaves are included in the wavefunction expansion,  
>>>>> hence the lower
>>>>> the energy. However in the ultrasoft pseudopotential definition  
>>>>> the energy
>>>>> is not only a function of the wavefunctions but also includes a  
>>>>> dependence
>>>>> on augmentation charges, that are localized and may contain  
>>>>> higher Fourier
>>>>> components with respect to 4*ecutwfc ( = 240 in your case). Failing to
>>>>> include enough Fourier components in the augmentation charges  
>>>>> will affect a
>>>>> number of integrals but not in a variational way... integrals  
>>>>> would simply
>>>>> be inaccurate and the inaccuracy can be both from above or from below.
>>>>>
>>>>>    - I would perform cutoff convergence test in a slightly different
>>>>> order: 1) I would check convergence of total energy (and stress, and
>>>>> forces) as a function of ecutwfc using the default value for
>>>>> ecutho=4*ecutwfc (that is without specifying ecutrho in the input). When
>>>>> this procedure converges (and it can initially converge from below due to
>>>>> augmentation charge Fourier components being missing) this means that
>>>>> wavefunction expansion AND augmentation-charge expansion are both
>>>>> converged. 2) I would then fix ecutrho=4*converged_ecutwfc, which takes
>>>>> care of augmentation charge convergence, and I would check whether I can
>>>>> get away with a lower ecutwfc for the wavefunction expansion.
>>>>>
>>>>>     - as for k-point sampling convergence, there is no variational
>>>>> principle w.r.t. number of k-points: it's again a matter of  
>>>>> convergence of
>>>>> an integral. The denser the grid the better the integral but there is no
>>>>> variational principle with respect to which k-point you include and which
>>>>> you dont.
>>>>>
>>>>>   hope this helps
>>>>>
>>>>> stefano
>>>>>
>>>>> On 24/04/2018 05:56, Martina Lessio wrote:
>>>>>
>>>>> Dear Quantum Espresso community,
>>>>>
>>>>> I am new to Quantum Espresso and I am trying to run some simple
>>>>> simulations on MoTe2 bulk. Unfortunately I seem to be having some issues
>>>>> with some preliminary convergence tests for charge density cutoff and
>>>>> K-point grid and I am hoping to get some help from you on this.
>>>>> Here is a graph with the results of the charge density cutoff
>>>>> convergence test I performed while setting the kinetic energy  
>>>>> cutoff equal
>>>>> to 60 Ry (I performed a test to set this as well):
>>>>>
>>>>> I am worried about these results because I would expect the total
>>>>> energy to go down rather than going up when I increase ecutrho. I also
>>>>> observe a similar energy trend when I increase the k-point grid,  
>>>>> which also
>>>>> seems unusual and possibly wrong to me.
>>>>> I am copying below the input I have used for these calculations and I
>>>>> would greatly appreciate any help with figuring our whether I am doing
>>>>> something wrong.
>>>>>
>>>>> Thank you so much!
>>>>>
>>>>> Kind Regards,
>>>>> Martina Lessio
>>>>>
>>>>> Postdoctoral Research Scientist
>>>>> Department of Chemistry
>>>>> Columbia University
>>>>>
>>>>>  &control
>>>>>
>>>>>     calculation = 'scf'
>>>>>
>>>>>     restart_mode='from_scratch',
>>>>>
>>>>>     prefix='MoTe2_ecutwfc',
>>>>>
>>>>>     pseudo_dir = '/home/mlessio/espresso-5.4.0/pseudo/',
>>>>>
>>>>>     outdir='/home/mlessio/espresso-5.4.0/tempdir/'
>>>>>
>>>>>  /
>>>>>
>>>>>  &system
>>>>>
>>>>>     ibrav= 4, A=3.530, B=3.530, C=13.882, cosAB=-0.5, cosAC=0, cosBC=0,
>>>>>
>>>>>     nat= 6, ntyp= 2,
>>>>>
>>>>>     ecutwfc =60.0 ecutrho=300.
>>>>>
>>>>>     nspin =4, lspinorb =.true., noncolin=.true.
>>>>>
>>>>>  /
>>>>>
>>>>>  &electrons
>>>>>
>>>>>     mixing_mode = 'plain'
>>>>>
>>>>>     mixing_beta = 0.7
>>>>>
>>>>>     conv_thr =  1.0d-8
>>>>>
>>>>>  /
>>>>>
>>>>> ATOMIC_SPECIES
>>>>>
>>>>>  Mo  95.96 Mo.rel-pbe-spn-rrkjus_psl.1.0.0.UPF
>>>>>
>>>>>  Te  127.6 Te.rel-pbe-n-rrkjus_psl.1.0.0.UPF
>>>>>
>>>>> ATOMIC_POSITIONS {crystal}
>>>>>
>>>>> Te     0.333333334         0.666666643         0.625000034
>>>>>
>>>>> Te     0.666666641         0.333333282         0.375000000
>>>>>
>>>>> Te     0.666666641         0.333333282         0.125000000
>>>>>
>>>>> Te     0.333333334         0.666666643         0.874999966
>>>>>
>>>>> Mo     0.333333334         0.666666643         0.250000000
>>>>>
>>>>> Mo     0.666666641         0.333333282         0.750000000
>>>>>
>>>>>
>>>>> K_POINTS {automatic}
>>>>>
>>>>>   8 8 2 0 0 0
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> users mailing  
>>>>> listusers at lists.quantum-espresso.orghttps://lists.quantum-espresso.org/mailman/listinfo/users
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> users mailing list
>>>>> users at lists.quantum-espresso.org
>>>>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Martina Lessio, Ph.D.
>>>> Frontiers of Science Lecturer in Discipline
>>>> Postdoctoral Research Scientist
>>>> Department of Chemistry
>>>> Columbia University
>>>>
>>>
>>>
>>>
>>> --
>>> Martina Lessio, Ph.D.
>>> Frontiers of Science Lecturer in Discipline
>>> Postdoctoral Research Scientist
>>> Department of Chemistry
>>> Columbia University
>>>
>>> _______________________________________________
>>> users mailing list
>>> users at lists.quantum-espresso.org
>>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>>
>>
>>
>> _______________________________________________
>> users mailing list
>> users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>
>
>
>
> --
> Martina Lessio, Ph.D.
> Frontiers of Science Lecturer in Discipline
> Postdoctoral Research Scientist
> Department of Chemistry
> Columbia University


GIUSEPPE MATTIOLI
CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
Via Salaria Km 29,300 - C.P. 10
I-00015 - Monterotondo Scalo (RM)
Mob (*preferred*) +39 373 7305625
Tel + 39 06 90672342 - Fax +39 06 90672316
E-mail: <giuseppe.mattioli at ism.cnr.it>



More information about the users mailing list