[Pw_forum] ecutvcut in Hybrid functionals

Simon Binnie sbinnie at sissa.it
Fri Nov 23 18:17:57 CET 2012


On Fri, 23 Nov 2012 17:41:19 +0100, Amin Torabi <mtorabi at uwo.ca> wrote:

> Dear Layla
>
> You said that "ecutvcut is a parameter which is not strictly bonded to
> hybrid functional parameters, but to strongly anisotropic supercells"
> So, why do we have ecutvcut=0.7 in the PBE0 example for bulk Si?
>
>

Hmm I think there seems to be a lot of confusion here. The first thing to  
understand is that the value ecutvcut has no effect if exxdiv_treatment is  
set to anything other than 'vcut_ws'.

By default all of the EXX examples run with exxdiv_treatment set to  
'gygi-baldereschi'. So in all these examples the value of ecutvcut is  
redundant.

However with extra command line options to run_example it is possible to  
run the examples with different exxdiv_treatment values. Hence why  
ecutvcut is set to some default value (0.7)

In the case of HSE however using vcut_ws (or vcut_spherical) is a rather  
dubious thing to do at best (due to the fact the functional already  
separates the exchange interaction into long and short range and only uses  
true non-local exchange in the short range part.)

As for the meaning of ecutvcut:

vcut_ws truncates the Coulomb potential *used in the exact-exchange  
integrals* in real space according to the extent of the Wigner-Seitz cell  
(taking into accoutn k-point sampling). The way it does this is by  
building the interaction on a real space grid and the fourier transforming  
it, right at the start of the calculation. This interaction (no in G  
space) is used throughout the rest of the calculation for the EXX terms.  
ecutvcut is the cutoff used in this transform and has to strike a balance  
between producing an accurate enough 1/|r-r'| function in G space and  
being so large as to take up too much time at the start of the calculation.

Brgds

Simon


-- 
Simon Binnie | Post Doc, Condensed Matter Sector
Scuola Internazionale di Studi Avanzati (SISSA)
Via Bonomea 256 | 34100 Trieste | sbinnie at sissa.it



More information about the users mailing list