[Pw_forum] Problems with phonones in Fe

Dragoni Daniele daniele.dragoni at epfl.ch
Mon Oct 28 15:39:08 CET 2013


1) You can parallelize on k points .. your cores are divided into pools of processors each taking care of a subset of k-points. Try to read some documentation by Prof. P. Giannozzi.

2) Have a look also at Shobhana Narasimhan DFPT work on iron. It depends on which quantity you are interested in. If you want high accuracy on the phonon dispersion, of course you can go even at denser grids. However, thermodynamic quantities are obtained via integration over frequencies in the reciprocal space, so that an approximate phonon dispersion is enough to get reliable thermodynamic results. You can try to check the free energy convergence w.r.t q-mesh (but it will take a bit).

3) You should always converge w.r.t your parameters according to the accuracy that YOU require. Again, try to converge the frequencies in a single q-point w.r.t this threshold.

            Daniele

__________________________________________

Daniele Francesco Dragoni
Laboratoire de théorie et simulation des matériaux
Swiss Federal Institute of Technology
EPFL STI IMX THEOS
MXC 318
Station 12
CH-1015, Lausanne



On Oct 28, 2013, at 10:24 AM, Иван Булдашев wrote:

Dear users, thanks for your replies, but I had some questions yet.
Firstly, with one Fe atom in BCC cell and 4x4x4 k-point grid my task
works faster, but it took two days. Are there any reasons for so long
calculation?

For iron on 48 cores I would probably parallelize on k-points using
npools, if the memory allows.

I do not use it. If I understand right, the key "-npool" can accelerate
calculation, because it divide process?

You are using a 8x8x8 grid in the cubic cell, do you really need it
for thermodynamic quantities?

My calculation was based on paper Tilmann Hickel and Joerg Neugebauer,
they used PHON (small displacement method) and took big grid. And I
can't find the same work with QE using. Can I get high accuracy with
4x4x4 grid?

As you see, even if it costs a bit more, I use a more strict
convergence threshold. Try to decrease your value.

I try 1e-16 and it works very long. Can I get same accuracy with
threshold = 1e-12?
_______________________________________________
Pw_forum mailing list
Pw_forum at pwscf.org<mailto:Pw_forum at pwscf.org>
http://pwscf.org/mailman/listinfo/pw_forum

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20131028/ab80a914/attachment.html>


More information about the users mailing list