The ABINIT Users Mailing List ( CLOSED )

[TORRENT Marc <marc.torrent@cea.fr>]

Dear Davide,

What do you mean by "are faster than abinit" ?
1- I suppose you did a convergence check on plane-wave cut-off... you 
should find an "ultrasoft" cut-off.
2- Having the same pw cut-off in all codes, Abinit can be slower, BY 
DEFAULT. Indeed, all defaults of Abinit are set in order to produce very 
accurate results (in the case you eventually want to perform linear 
response calculations with your ground state wavefunctions). If you want 
Abinit to run faster you should decrease some parameters (some are 
tricky, like, for instance boxcutmin...). Especially for FFT grids which 
are very large with Abinit (play with ngfft, ngfftdg, ...)
3- I also suppose that you took advantage of all the capabilities 
concerning parallelism (k, fft, bands).

If Abinit is still slower... you can only hope that some developper 
wants to optimize it.

Concerning spin-polarized systems, yes Abinit+PAW is compatible: some 
provided PAW datasets can contain a non-spin-polarized compatible 
"pspxc" keyword; but this is easy to bypass: simply replace pspxc=2 by 
pspxc=7 in your PAW dataset file; or put ixc=7 in your input file...

Marc Torrent
CEA - Bruyeres-le-Chatel - France


Davide Sangalli a écrit :
> Dear forum,
>    I'm trying to compare the speed of abinit versus some
> other DFT code and I realized that other codes which uses
> ultra-soft pseudo-potential are faster than abinit.
>
> Anyway, from what I undestood, using PAW one should be
> able to have performances like "DFT ultrasoft codes"
> if the starting point for the computation is a converted ultrasoft.
>
> I tryed to use the PAW atomic data table from the abinit website:
> http://www.abinit.org/PAW/MAIN/PAWTable.html
> but unfortunately they are not suited for spin polarized calculations.
>
> So two questions:
>  - Can I use PAW for spin polarized calculations?
>  - Using PAW are there particularly drawbacks with respect to the
>     standard pseudo-potential methods?
>
> Thanks. Best regards,
> Davide