[QE-users] higher total energy in the last stage of vc-relax

[Kevin May]

Hi Mehrdad,

Just a few more thoughts:

1) Which reference state are you using for Fe? Jiang et al. use ferromagnetic
BCC Fe in Acta Materialia 56, 3236–3244 (2008). This paper uses constant
volume rather than constant (zero) pressure calculations. I would
definitely do spin polarized calculations. You mentioned your current
pseudopotentials are not recommended for spin polarized calculations--I
would use different ones.

2) I'm not sure which pseudopotentials you are using exactly, but I have
had good results using those in the Standard Solid State Pseudopotentials
(SSSP) tables, which comes from very useful work of a team at EPFL (
https://www.materialscloud.org/discover/sssp/). In your case, this would be
from PSLibrary 0.3.1 PAW for Fe and PSLibrary 1.0.0 PAW for C. Note the
recommended ecutrho is 12*ecutwfc for Fe. The recommended cutoff there is
90 Ry but you might get away with something smaller if you test it yourself.

3) A good reference for defect calculations is Rev. Mod. Phys. 86, 253
(2014).

Best,

Kevin May, PhD
Postdoctoral Associate
Department of Materials Science and Engineering
Massachusetts Institute of Technology




On Thu, May 30, 2019 at 3:59 AM mehrdad zamzamian <
mehrdad.zamzamian at gmail.com> wrote:

> Thanks for your attention to my question
> Actually, i defined a supercell (2*2*2), used USPP-PBE.I adjusted the
> cutoff according to what was proposed for it (~70) and for ecut_rho (~400).
> for scf convergence, i set conv=1e-6, ant at here, i set mixing beta~0.04
> (it helped to reach to convergence fast). and also i set press=0 and its
> threshold to 0.5. I want to calculate the vacancy (lack of Fe) energy. its
> value must be 1.6 eV but i never give the better than 2.2 eV (The
> ridiculous thing is that with molecular dynamics I got a very precise
> amount of 1.6! that of course, its interatomic is more precise than my
> calculation!). according to my PP in this calculation, it is suggested to
> use non-polarized spin. i really do not know what parameter can i changed
> to achieve 1.6 eV.
> In addition, your food for the brain is some hard to be digested!
>
> Regards
>
> Seyed Mehrdad Zamzamian
> Sharif University of Technology, Tehran, Iran
> Energy engineering department
> E-mail: mehrdad.zamzamian at gmail.com
>
> On Wed, May 29, 2019 at 9:48 PM Kevin May <kmay at mit.edu> wrote:
>
>> Hi Mehrdad,
>>
>> If you have a large energy difference between the end of the vc-relax
>> algorithm and the final scf step, you may be using a fairly low plane wave
>> cutoff energy for your system, though that may be fine depending on what
>> quantities you are looking at. An important question here is: how are you
>> choosing your convergence criteria? In my experience stresses (vc-relax)
>> require a higher cutoff compared to total energy or forces (just ionic
>> relaxation).
>>
>> This all depends of course on what you are trying to get out of your
>> calculation. In my case I was comparing different magnetic ground states
>> that could be very close in energy, where very small changes in geometry
>> can make a difference. For my specific systems I would converge plane wave
>> cutoff, k-points and convergence threshold with respect to unit cell stress
>> using a reference calculation with very high cutoff, low threshold (10^-9
>> Ry), and dense k-point mesh.
>>
>> Just some food for thought. I've definitely seen papers in the literature
>> where they claim state A has lower energy than state B. I've reproduced
>> such results using the somewhat lax cutoffs reported, and then found when
>> you actually do CONVERGED calculations, state B is actually lower in energy
>> (whoops!). Convergence is important.
>>
>> Best,
>>
>> Kevin May, PhD
>> Postdoctoral Associate
>> Department of Materials Science and Engineering
>> Massachusetts Institute of Technology
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190530/936f57dc/attachment.html>