[Pw_forum] PBE0 Hybrid functional SCF doesn't complete

[Tiana Davide]

Hi Thierry,
I don't know if it's only for the system I am working on or general but:

I found in the last version of PW, 5.4, the hybrid are much more faster so you could upgrade it
Consider of using HSE06 instead of PBE0 

Cheers
Davide
________________________________________

Message: 1
Date: Thu, 23 Jun 2016 12:32:06 +0000
From: Thierry Clette <thierry.clette at hotmail.com>
Subject: Re: [Pw_forum] PBE0 Hybrid functional SCF doesn't complete
To: PWSCF Forum <pw_forum at pwscf.org>
Message-ID:
        <DB5PR1001MB09332DBD243D232ED7DDAB4C872D0 at DB5PR1001MB0933.EURPRD10.PROD.OUTLOOK.COM>

Content-Type: text/plain; charset="windows-1252"

So that is how I did it wrong


Thanks for the tips


Have great day


Thierry Clette

________________________________
De : pw_forum-bounces at pwscf.org <pw_forum-bounces at pwscf.org> de la part de stefano de gironcoli <degironc at sissa.it>
Envoy? : jeudi 23 juin 2016 10:33:17
? : PWSCF Forum
Objet : Re: [Pw_forum] PBE0 Hybrid functional SCF doesn't complete

the Hybrid calculations are VERY expensive.
in solids the expensive Fock step has an inner summation on k-point that makes it QUADRATIC in the grid dimension used (20*20*20)^2, in your case.
No wonder it does not complete

Things I would do:
- reduce the cutoff to the minimum.
- consider using a NC pseudopential.
- please no gaussian smearing. use 'mv' or 'mp'.
- use larger smearing => smaller k-points set
- read, understand and test the nqx1,nqx2,nqx3 input options
- look for a big machine and use band parallelization to distribute work load.

HTH

stefano

On 22/06/2016 12:22, Thierry Clette wrote:

Hello,


I'm a student at the University of Brussels. I use QE 5.2.1.

I'm trying to do a SCF calculation of bulk black Phosphorus, a semiconductor with a small gap. PBE0 seems to be the right method for this, so I use a normcon PBE pseudopotential from the QE site with the option "input-DFT=PBE0". The inputs and outputs are attached.

As you can see in the output, the calculation in itself seems to work, but the process doesn't seem to finish. It seems it can't write to the save. Usually the process never finishes and is killed by time out. In some cases, it is killed for using too much RAM (after the calculation is complete, very strange). Anyway, as it doesn't finish, I can't do band calculations.

Are some of my parameters wrong ?


Thanks in advance


Thierry Clette
________________________________