[DFTB-Plus-User] Forces in dftb+ (supercells and k-point mesh)

Anshu Gaur agaur at iitk.ac.in
Sat Mar 21 08:24:51 CET 2020


Dear Bálint,

Thanks for your reply.
My question is related to article " Towards a simplified description of thermoelectric materials: accuracy of approximate density functional theory for phonon dispersions", J. Phys.: Condens. Matter 31 (2019) 395901. In this paper structures are relaxed to very small force components (individual components in x- y- and z- directions on every atom) and then Phonopy module for dftb+ is used to calculate phonon dispersion with single k-point "Gamma" calculations for distorted super-cells for force calculations.

I am trying similar approach in my calculations. Hoever, the question remains that if I make a super-cell (and use dftb+ for force calculations with same k-point grid) or use single k-point "Gamma" for force calculations, I see large force components on individual atoms. In such a case, would the force calculations on distorted super-cell and hence the resulting phonon dispersion be correct? The "Drift forces" which are subtracted during calculation of "FORCE_SETS" in phonopy calculation remains zero, but the "Drift forces" is the sum (vector) of forces on all atoms (in undistorted super-cell) rather than force on individual atoms (which are non-zero). The resulting phonon dispersion curves are very reasonable.
However, in my calculations, I am focusing on pre-existing forces on individual atoms before Phonopy calculations are carried out, as I need to know how phonon-modes and phonon dispersion is affected by presence of these forces (intentionally introduced). Does the approach in the above mentioned article still valid, even if there are some forces on atoms (coulombic, stress etc.) in the original structure?
Am I missing something here?
Thanks and regards,
Anshu Gaur

Date: Wed, 18 Mar 2020 07:16:03 +0100
From: Bálint Aradi <aradi at uni-bremen.de>
To: dftb-plus-user at mailman.zfn.uni-bremen.de
Subject: Re: [DFTB-Plus-User] Forces in dftb+ (supercells and k-point mesh)
Message-ID: <21f258be-af61-22c0-d43d-bfc491ef6928 at uni-bremen.de>
Content-Type: text/plain; charset="utf-8"

Dear Anshu,
As with all method (DFTB, DFT, etc.), you can only trust forces, if your
k-point mesh is convergent. You can test this by increasing your k-point
sampling systematically and check how the forces change (as you have
already done). In some specific 2D materials (e.g. graphene) it may
require a lot of k-points to get forces converged at low electronic
temperatures.

Otherwise, please not that 1e-6 Hartree/Bohr is in my opinion a very
tight relaxation criterion, 1e-4 is usually fine enough for most purposes.

Best regards,
Bálint

On 16.03.20 08:45, Anshu Gaur via DFTB-Plus-User wrote:
>
> Dear all,
>
> I am trying to see effect of forces (of different origins such as
> coulombic, van-der walls, strain etc.) on the properties of a structure.
> I have relaxed the structure (< 1e-6 Hartree/Bohr) using a sufficiently
> large k-point mesh. However, if I change the k-point mesh size (smaller)
> or make a supercell of the original structure, I see large forces in the
> structure (without changing atomic positions or lattice parameters).
> Also, I notice that the x-component of forces on atoms is the most
> affected (while y- and z-components remain unaffected). I have checked
> this by changing k-mesh in x-, y- and z- and also making supercells in
> x- and z- directions, the x-component of the forces always shows large
> values.
> Can the calculation of forces at any other k-point mesh be trusted, if
> the structure is relaxed using a different k-point mesh? or we should
> always use the same k-point mesh. Also, can we make a supercell and
> trust the force calculations?
>
> My structure is 2D with vacuum in y- and extended in x- direction
> (ribbon like structure, continuous in x- and z-).
>
> Regards,
> Anshu
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20200321/a7065e20/attachment.htm>


More information about the DFTB-Plus-User mailing list