[Pw_forum] questions about vc-relax or geomtry

[Stefano de Gironcoli]

  the variable cell shape MD part in PWscf is based on Renata Wentzcovitch's
code, described in R.M. Wentzcovitch, PRB 44, 2358 (1991).
  Parrinello-Ramhan MD is also available from the same code.
  Notice however that the meaning of cell-mass in the two formulations is 
not the same. For a discussion of these issues one should look at the 
original references.
   The default value in PWscf are set to
   wmass=0.75*Tot_Mass/pi**2 for Parrinello Rahman MD and to
   wmass=0.75*Tot_Mass/pi**2/Omega**(2/3) for Wentzcovitch MD.
   Documentation should be corrected.

   I think that it is a good idea to start from the default value and 
possibly tweak a bit around it.
   I think it is a good idea to relax the internal coordinates before
starting the  variable cell-shape optimization.
  I think that it is good policy to set all atomic masses to the same
value (for instance 1.0) if one is interested in relaxation so as to let
Hydrogen and Uranium react in the same way to the forces, as the 
variation in interatomic forces is way less significant than (and 
uncorrelated with) the variation of atomic masses.
   I think that one should gain some idea what the proper time step
should be with some short MD (undamped) run to estimate the period of
free oscilaltions.
   Similar systems are likely to need similar time-steps.

   The damped dynamics performs a quickmin-style damping: at each step
the velocity (of the internal coordinates or of the cell) is projected on 
the direction of the forces (and stress). When they happen to have a 
negative projection the corresponding velocity is stopped. No damping
factor is needed.

   Hope this helps,

    stefano

PS: the part of the code dealing with cell dynamincs is fairly 
independent from the rest of the code: only some information
on the current and previous atomic&cell positions and forces&stress
values are needed at each iteration.
Whoever has the perfect, plug-and-play, easy-to-use, robust, efficient, 
well documented variable-cell-shape enthalpy minimizer is most welcome to 
provide it to the community. It will be relatively easy to add this 
functionality. thanks.


On Mon, 20 Aug 2007, Eduardo Ariel Menendez P wrote:
>
> Hi,
> I take the occasion to ask some clarifying on the variable-cell
> optimization. Maybe it seems difficult because is not well documented.
> After some laborious vc-relax calculations with small unit
> cells, the optimal values of wmass are still misterious for me.
>
> When I follows the the instructions in INPUT_PW
>
> wmass          REAL ( default = sum of atomic masses in the cell )
>               fictitious cell mass [amu] for variable-cell simulations
>               (both 'vc-md' and 'vc-relax')
>
> then I see that the cell varies extremely slowly. Then I have chosen
> "sum of atomic masses in the cell"/Factor using Factor as 100,1000,etc
> until I see and effective change in volume. When it happens I see the volume
> converges fast, but then the shape began to fluctuate for a long time
> (maybe this is the moment to use Factor=1).
> If wmass is not specified pw provides a value that
> is reported in the output and is much smaller than "sum of atomic masses
> in the cell".
>
> For example, a Pt in HCP structure, with two atoms per cell as
> "sum of atomic masses in the cell"=2x195.084=390.168
>
> the default reported by pw when wmass is unset is "0.00300 AMU/(a.u.)^2"
> There is a factor of 130000, which I do not understand. ¿What is a.u. in
> AMU/(a.u.)^2 ?
>
> The units of W seems to be E*T^2/V^2 (E=Energy,V=Volume,T=time) in the
> Andersen and Parrinello-Ramman Lagrangians.
> http://www.vlab.msi.umn.edu/events/download/vlab_lectures/renata/lecture1.pdf
> Is the factor related to the volume?
> In my case of Platinum el volumen es 203.4(a.u.)^3., so V^2=41371 which is
> near 130000/3.
>
> Hence, some relation is betwwen the mass in AMU, the volume, and maybe the
> time step. ?Can someone explain it?
>
> Second, it seems that one value is appropriate for volume relaxation, and
> other for shape relaxation, depending on the elastic constants. I
> welcome  know-how sharing.
>
> Third. In damped dynamics, there should be some damping factor, but it
> seems to be fixed in PWSCF. Is it?
>
> Eduardo A. Menendez Proupin
> Department of Physics
> Faculty of Science
> University of Chile
> Las Palmeras 3425
> Ñuñoa, Santiago
> Chile
> Phone: (56)(2) 978 74 11
> Fax  : (56)(2) 271 29 73
> http://fisica.ciencias.uchile.cl/~emenendez/
>
>
>
>
> ? 2007-08-19?? 20:44 -0400?Axel Kohlmeyer???
>> On Mon, 20 Aug 2007, jideson wrote:
>>
>> J> Dear experts,
>>
>> dear jideson,
>>
>> J>      I read some  emails about vc-relax in this list, but  still have
>> J> some questions.  It seems that  geometry optimization (both cell and
>> J> atomic postions) in pwscf is a diffcult job than other softwares such
> as
>>
>> i disagree. it is not difficult, you just need to know what you are
>> doing and particularly know what you can expect. the quantum espresso
>> package allows you a lot of flexibility, but the price for that is
>> that not everything works automatically right. usually, if something
>> is not working easily, it is an indication, that you have not yet
>> understood what you are doing (or else it would be obvious) or
>> you have a problem that is not easy to start with or (which is rare)
>> have found a bug in the code.
>>
>> J> CASTEP.  Can you give me a step-by-step tutorial ?
>>
>> there are some input examples within the distribution
>> and you may want to check out material from previous QE
>> tutorials, e.g., the material at:
>> http://www.vlab.msi.umn.edu/events/lecture.shtml
>>
>> J>    By the way , another question , are there differences between cp.x
>> J> and cpmd(www.cpmd.org) code ?  Thanks!
>>
>> yes. both have the same origin, there have evolved into
>> quite different species by now.
>>
>> cheers,
>>    axel.
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>