Problem in the ALGO = CHI step.

Question on input files/tags, interpreting output, etc.

Please check whether the answer to your question is given in the VASP online manual or has been discussed in this forum previously!

Moderators: Global Moderator, Moderator

Post Reply
Message
Author
Bikash_Patra
Newbie
Newbie
Posts: 2
Joined: Tue Jan 31, 2017 12:36 pm

Problem in the ALGO = CHI step.

#1 Post by Bikash_Patra » Thu May 13, 2021 11:42 am

Dear users,
I want to calculate the optical properties of my system.
For this, I performed three calculations first with ALGO = normal,
then with ALGO = Exact (ISTART = 1) and third with ALGO = chi.
First two steps completed successfully but I am getting some error
in the third step. I have attached all the necessary files.
You do not have the required permissions to view the files attached to this post.

henrique_miranda
Global Moderator
Global Moderator
Posts: 194
Joined: Mon Nov 04, 2019 12:41 pm
Contact:

Re: Problem in the ALGO = CHI step.

#2 Post by henrique_miranda » Fri May 21, 2021 2:29 pm

I want to calculate the optical properties of my system.
Firstly you should clarify which optical properties and under which approximation do you need.

From this tutorial (that I guess you had a look at):
wiki/index.php/Dielectric_properties_of_SiC

You can compute the static dielectric function or the frequency-dependent dielectric function with or without local field effects (independent particle approximation).
It might be worth (in case you did not do so already) to first compute the frequency-dependent dielectric function in the IPA:
wiki/index.php/Dielectric_properties_of ... le_picture

In step 2 you are using exact diagonalization (ALGO=Exact) for a fairly large system (56 ions).
Besides requiring a large amount of memory it takes a lot of time to compute.
If you are only interested in the imaginary part of the dielectric function near the gap or fermi energy, there is no need to use exact diagonalization and compute so many bands.
You can set ALGO=Normal and include a few conduction bands.

Furthermore, when using ALGO=Exact and provided you start from a pre-converged calculation (i.e. the WAVECAR from step 1 is present or the CHGCAR file is present and ICHARG=1 is set in the INCAR file), you should set NELM=1 which tells VASP that it should do only one step of electronic minimization.
In your case, the exact diagonalization was done twice which is wasteful (your calculation of step 2 took 2 days, it would have taken only 1 day instead).

In step 3 the problem is the memory usage.
in particular, in the last line of your OUTCAR you can see:

Code: Select all

 total amount of memory used by VASP MPI-rank0 73564944. kBytes
=======================================================================

   base      :      30000. kBytes
   nonl-proj :     280960. kBytes
   fftplans  :       3830. kBytes
   grid      :      27462. kBytes
   one-center:        870. kBytes
   HF        :        259. kBytes
   nonlr-proj:      11432. kBytes
   wavefun   :      76896. kBytes
   response  :   73133235. kBytes
so you trying to use using 73GB of memory per rank.
There seems to be an issue with the reporting of the memory usage in VASP and that is why you did not get an error message warning you that that is the issue.
We will fix this in a future release.

If you really need to perform this calculation including local-field effects, you should look at these variables to reduce the memory usage of your calculation:
wiki/index.php/ENCUTGW
wiki/index.php/LSPECTRAL
I recommend that you first try to run these calculations on smaller systems and look at the memory usage and only then attempt the larger calculation you are attempting to do now.

Post Reply