[Pw_forum] relax question

[Stefano Baroni]

Hi All:

a small further comment to explicitly stress a point that has already  
been raised more or less implicitly.

One runs simulations not just for the sake of it, but to calculate  
properties, right? And properties are numbers, right?

Well, the right questions to ask at the start of a simulation concern  
the influence of the input parameters on the *specific numbers* one  
wants to calculate (i.e. the *specific property* one wants to study).  
Of course, the energy is a very important property, and it is very  
correct to ask oneself what the influence of the input parameters on  
the accuracy of the calculated energy would be (not to mention its  
conservation, if one runs a MD simulation). It is not the only one,  
though. If you want to study the equilibrium geometry, for instance,  
it is the accuracy of bond lengths and bond angles that would matter,  
and the softer the system, the more difficult it will be to achieve a  
given target accuracy (this point has been discussed in this very  
thread, I think).

All in all, I would recommend that before asking what value a given  
parameter should be given, one should ask oneself "why should I care  
about this parameter?". A proper answer to the latter question will  
almost invariably also answer the first.

I understand that the usage of canned packages such as QE or vasp  
encourages a somewhat uncritical approach to the many, many,  
parameters needed to fine tune a well done simulation. This is one  
more reason, I think, to make an effort to ask sometimes general  
questions before specific ones.

Cheers - Stefano


On Jul 15, 2007, at 5:29 AM, Xunlei Ding wrote:

> Hi,
>
> In the vasp forum, one set the parameter like this:
> EDIFF: 1E-07
> EDIFFG: -0.0005
>
> And in the reply it is said, "Anyways, your ediffg is very low. A  
> strict force
> criteria is -0.001, and many people would use -0.01 or higher. If  
> you look at
> the change in the energy as you go from a force of 0.01 eV/Ang to  
> 0.001 eV/Ang,
> it is usually negligible."
>
> You can check the difference of results when you set force_conv_thr  
> as 1.d-3 and
> 1.d-4. If the difference is negligible, why use too strict threshold?
>
> Yours,
> Ding Xunlei
>
>
> Quoting xu yuehua <njuxuyuehua at gmail.com>:

---
Stefano Baroni - SISSA  &  DEMOCRITOS National Simulation Center -  
Trieste
[+39] 040 3787 406 (tel) -528 (fax) / stefanobaroni (skype)

Please, if possible, don't  send me MS Word or PowerPoint attachments
Why? See:  http://www.gnu.org/philosophy/no-word-attachments.html



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20070715/65291560/attachment.html>