[QE-developers] Significant Deviations from Elastic Tensors from existing codes

Wed Feb 13 08:14:54 CET 2019

Dear Christopher,

A few comments.

- Which pseudo are you using? Also, the cutoff might need to be increased (I imagine what you state is the wfc cutoff, what about the charge density cutoff?)
E.g. in the link Nicola sent, they use 56/219 as cutoffs, with SSSP pseudos https://molmod.ugent.be/sites/default/files/deltadftcodes/supplmat/SupplMat-QE-SSSP-acc.pdf
And also the k-mesh density they use is of 20x20x20 (if they used it probably it’s a good indication you might need to go to that convergence to ensure that you results are correct)

- more generally, it would be easier and you wold get faster feedback if you could share your calculation inputs and outputs, and a detail of how you computed the constants

- in the link that Nicola provided, if you check the PDF for the “best" QE and VASP calculations (the one with smaller delta w.r.t. a full-electron code, wien2k in this case), you will find the values of B0 and B1, namely for Silicon

B0=88.698 GPa, B1=4.318 [QE, link above]
B0=88.753 GPa, B1=4.297 [VASP: https://molmod.ugent.be/sites/default/files/deltadftcodes/supplmat/SupplMat-VASP2015GW.pdf]

This shows that VASP and QE are in very good agreement.
Also, considering that for cubic systems B0 = (c11 + 2*c12)/3, if I’m not mistaken from the numbers you report you would get (for B0) 83.3 for VASP, 93.3 for QE.
As I tend to believe that the one published at the links are converged, I think this means that both your reference data from the Materials Project and your simulations still are not fully converged. (As a note in passing, this also shows that we shouldn’t always believe that all data from online databases is always well converged - especially if you cannot check the exact inputs and input parameters that they used.)

If you redo the simulations with more converged parameters and you still have discrepancies (from the values from the molmod link), I suggest that you send us the calculation details (as I mention above) so that we can investigate more.

Best,
Giovanni

--
Giovanni Pizzi
Theory and Simulation of Materials and MARVEL, EPFL
http://people.epfl.ch/giovanni.pizzi
http://nccr-marvel.ch/en/people/profile/giovanni-pizzi

On 12 Feb 2019, at 22:28, Christopher Pashartis (----) <Christopher.Pashartis at imec.be<mailto:Christopher.Pashartis at imec.be>> wrote:

Hello Nicola,

I am aware of many of the items you have stated. If you could show me a calculation of the Elastic Tensor with ESPRESSO for Si I would be more inclined to believe that ESPRESSO is not at fault.  I tested different pseudopotentials, it didn't help ESPRESSO's case. I did converge to what I thought was an appropriate cutoff and kpoint mesh (I believe I used 30Ry and 11x11x11, comparable to materials project parameters used in VASP).

Again, perhaps just a thought from me, it could be due to breaking the symmetry of the crystal as is the case when you apply a deformation.

I will get back to you as soon as I can with more information. On the meantime, it might be nice if one of you can do a calculation with ElaStic to what you may believe would be overconverged values.

P.S. Unless I am missing the point, I fail to see how the link you showed me affects the elastic tensor.
________________________________
From: Nicola Marzari <nicola.marzari at epfl.ch<mailto:nicola.marzari at epfl.ch>>
Sent: Tuesday, February 12, 2019 9:18 PM
To: General discussion list for Quantum ESPRESSO developers; Christopher Pashartis (----)
Subject: Re: [QE-developers] Significant Deviations from Elastic Tensors from existing codes

Dear Christopher,

thanks for reporting, but no bug - QE and VASP (if you use GW-ready
pseudo) have actually the best agreement of bulk moduli with all-
electron codes: https://molmod.ugent.be/deltacodesdft

Discrepancies between codes are due to different pseudopotentials or the
numerically details needed to get a well converged calculation - not
trivial, at al:  it depends if you are fitting the energy or the stress,
how much strain you are using, superconvergence on cutoffs, kpoints and
selfconsistency - hence why most values in the literature are poorly
converged, and properly done calculations, as those in the link above,
agree very closely.

                                nicola

On 12/02/2019 07:49, Christopher Pashartis (----) wrote:
> Dear Developer(s),
>
> I have encountered what would appear to be a serious bug in your code. I
> have been trying to tabulate some elastic constants for Si and Cu only
> to realize that your code has some issues with some components of the
> elastic tensor.
>
> Namely, the c44 component of Si and c11 component of Cu. All my other
> parameters are within a reasonable range of existing VASP calculations.
>
> For example Si (in GPa):
> Component
>        Materials Project (VASP)
>        ESPRESSO with the same routine as Materials Project (I did these calcs)
> c11
>        144
>        156
> c12
>        53
>        62
> c44
>        75
>        104
>
> I have converged my cutoff parameters and kpoint mesh. I also tried
> fiddling with the tolerances on energy and forces./*All of these are
> converged in the values I show you above*/.It would seem both your
> energies and forces are miscalculated in certain deformation directions
> (and no I did not apply too much of a deformation). I am fairly certain
> that these issues have nothing to do with my code as I also implemented
> it for CP2k and the results were within acceptable margins of existing code.
>
> If you require any more information to assist you in this effort, let me
> know.
>
> *CHRISTOPHER PASHARTIS*
>
> PhD Student in Computational Materials
>
> christopher.pashartis at imec.be<mailto:christopher.pashartis at imec.be><mailto:christopher.pashartis at imec.be>I_www.imec.be_<http://www.imec.be/>
>
> Kapeldreef 75I3001 LeuvenIBelgium
>
> __<https://www.facebook.com/imecinternational/>__<https://twitter.com/imec_int>__<https://www.linkedin.com/company/imec?trk=company_logo>__<https://vimeo.com/imecnanotube>__<https://www.youtube.com/user/imecnanotube>__<https://www.flickr.com/photos/imec_int<https://www.facebook.com/imecinternational/%3E__%3Chttps://twitter.com/imec_int%3E__%3Chttps://www.linkedin.com/company/imec?trk=company_logo%3E__%3Chttps://vimeo.com/imecnanotube%3E__%3Chttps://www.youtube.com/user/imecnanotube%3E__%3Chttps://www.flickr.com/photos/imec_int>>
>
> imec e-mail disclaimer<http://www2.imec.be/be_en/email-disclaimer.html>__
>
> <http://www.imec.be/>
>
>
> _______________________________________________
> developers mailing list
> developers at lists.quantum-espresso.org<mailto:developers at lists.quantum-espresso.org>
> https://lists.quantum-espresso.org/mailman/listinfo/developers
>

--
----------------------------------------------------------------------
Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL
Director, National Centre for Competence in Research NCCR MARVEL, EPFL
http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project
_______________________________________________
developers mailing list
developers at lists.quantum-espresso.org<mailto:developers at lists.quantum-espresso.org>
https://lists.quantum-espresso.org/mailman/listinfo/developers

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20190213/efdfa24e/attachment-0001.html>