[Wien] wien.struct <--> abinit.in

[Steven Homolya]

On Wed, 24 Sep 2003, Jorissen Kevin wrote:

> I almost got to using abinit for structure optimization, in particular
> because it should be able to optimize lattice parameters and internal
> parameters at the same time, which is not possible in WIEN.  Almost, so
> I can't help you here - I just wanted to ask an additional question :
> How about calculation time?  How fast is an abinit-optimization?
> 

It is order(s) of magnitudes faster for small cases (say 1--10 atoms)  
because data manipulation and k-point parallelisation are very efficient
(it's all coded in fortran, no scripts, efficient file i/o). E.g., you can 
use 64 CPUs for 64 k-points and get there 63+ times quicker than on one 
CPU.

It is considerably faster for largish cases (20+ atoms), especially
because unit cell geometry and internal parameters need far fewer
converged SCF calculations than in wien. (Geometry optimiser seems quite 
smart.)

As far as time per single SCF cycle is concerned it's similar to wien if
one uses the same k-mesh and 'good' K.E. cutoffs (though usually you get
away with fewer k-points in abinit). But the SCF calc. is quicker to
converge with the psp & plane wave basis than with the full-electron
hybrid basis in wien. Also, as yet I haven't had to mess around with
mixing parameters; the program seems to guide convergence well.

Unfortunately, only k-point parallelisation is available for ground state
calcs., so you're stuck with 1 CPU if you've got a large unit cell. I
haven't checked out response function calcs. yet.

Steve

-- 
Steven Homolya
School of Physics and Materials Engineering
Monash University, VIC 3800
Australia
Tel: +61 3 9905 3694
Fax: +61 3 9905 3637