[Pw_forum] vc-relax : total force does not converge (Lorenzo Paulatto)

[Jiseok Kim]

Dear, Dr. Lorenzo Paulatto

Thank you very much Dr. Paulatto.
I'm learning the DFT by myself so there's still lots of things that I have
to learn.

Regarding the 'forc_conv_thr', I agree with you.
Originally I set it to 10^-4 and it converged with both 'relax' and
'vc-relax'.
But what happened is that I got negative frequencies (spaghetti-like over
the 1D BZ) when I calculate the phonon dispersion.
So I decided to go back to the geometry optimization and pushed the
'forc_conv_thr' even lower.
I admit that I fail to notice if the 10^6 was too low precision..


1. why are you using "cg" diagonalization? If the electronic calculation
does not converge with the default algorithm, there is probably a problem
in your input; changing algorithm will just mask it.

=> I don't think I have a problem with 'cg' diagonalization..
     It converges mostly within 10 iterations..



2. why are you mixing ultrasoft (Si) and norm-conserving (H)
pseudopotentials? It is not forbidden, but you get the worst of both
methods: high wfc cutoff (NC) and high charge density cutoff (US).


=>  I'm using 'Si.pbe-rrkj.UPF' and 'H.pbe-vbc.UPF'.
      As long as I understand, both pseudopotentials are norm-conserving..
      Please correct me if I'm wrong.
      I appreciate your point regarding the mixing that I haven't thought
of.


3. does it make sense to do vc-relax in a isolated 1D system? Can't the
system just rotate if it wants more space?


=> I very appreciate that you point this out.
     I want to learn and hear some advice whether the 'relax' or
'vc-relax' is more appropriate for the isolated system (1D and 2D)..
     As I've done, the 'relax' does not dramatically modify the geometry.
     There's only tiny differences in bonding lengths and bonding angles ..
     As I mentioned, I decided to 'vc-relax' because of the negative
frequencies of phonon dispersion..
     (I already try to decrease the 'tr2_ph' and 'conv_thr' when doing
ph.x)
     I'm not sure if the negative frequency indicates the 'real'
instability or not..
     But as far as I understand from literatures this geometry is supposed
to be stable..
     Also, I'm not sure if the 'vc-relax' can fix the problem..


Thank you very much again.
If there is anything that I'm thinking wrong, please correct me.

Sincerely,

Jiseok Kim



>
>Message: 2
>Date: Wed, 5 Sep 2012 12:06:18 +0200
>From: Lorenzo Paulatto <lorenzo.paulatto at impmc.upmc.fr>
>Subject: Re: [Pw_forum] vc-relax : total force does not converge
>To: PWSCF Forum <pw_forum at pwscf.org>
>Cc: rameshchitumalla at gmail.com,
>    CAO1fJnvo6EhnZUdMYhURsg2Y9fDQimng16PYiqLUOUCgfCXFFw at mail.gmail.com
>Message-ID:
>    <CAG+GtJdxAGxYGUVOygFNz532-vwK4V55U_u-6r0=R6GE2JV+wg at mail.gmail.com>
>Content-Type: text/plain; charset="iso-8859-1"
>
>On 5 September 2012 10:22, Jiseok Kim <tube77 at gmail.com> wrote:
>
>> Yes. I checked the geometry before and after the 'relax' which looks ok
>>...
>> The atomic distance between neighboring atoms about 12A along
>>x-direction
>> and about 24A along y-direction..
>> The vacuum region along x-direction might not be large enough but I
>> believe it wouldn't be a problem having considered that the 10~15A of
>> vacuum region usually is accepted. (Please correct me if I'm wrong.)
>>
>
>I think your forc_conv_thr is just too low, if you want to reach that
>ridiculous precision you need to reduce the conv_thr and/or increase
>upscale. I have a couple more observations:
>
>1. why are you using "cg" diagonalization? If the electronic calculation
>does not converge with the default algorithm, there is probably a problem
>in your input; changing algorithm will just mask it.
>2. why are you mixing ultrasoft (Si) and norm-conserving (H)
>pseudopotentials? It is not forbidden, but you get the worst of both
>methods: high wfc cutoff (NC) and high charge density cutoff (US).
>3. does it make sense to do vc-relax in a isolated 1D system? Can't the
>system just rotate if it wants more space?
>
>bests regards, and good work
>
>-- 
>Dr. Lorenzo Paulatto
>IdR @ IMPMC -- CNRS & Universit? Paris 6
>phone: +33 (0)1 44275 084 / skype: paulatz
>www:   http://www-int.impmc.upmc.fr/~paulatto/
>mail:  23-24/4?16 Bo?te courrier 115, 4 place Jussieu 75252 Paris C?dex 05
>-------------- next part --------------
>An HTML attachment was scrubbed...
>URL: 
>http://www.democritos.it/pipermail/pw_forum/attachments/20120905/79086103/
>attachment-0001.htm
>