The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

Dear Ane Xu,

Ane Xu wrote:
> Dear all:
>     I am conducting structure optimization on a unitcell of carbon
> nanotube, using following parameters (ecut and k-points are optimized
> for energy), but in my log file I found these problems.
> 1) 
> /*
>   getcut: wavevector=  0.0000  0.0000  0.0000  ngfft=  90  90  54
>          ecut(hartree)=     35.000   => boxcut(ratio)=   2.41416
> 
>  getcut : COMMENT -
>   Note that boxcut > 2.2 ; recall that boxcut=Gcut(box)/Gcut(sphere) = 2
>   is sufficient for exact treatment of convolution.
>   Such a large boxcut is a waste : you could raise ecut
>   e.g. ecut=   50.996504 Hartrees makes boxcut=2
> */
> 
>   But 500eV is enough in some reference papers(using VASP, Castep...),
> so I wonder how can I do to correct this "waste" ? Raise ecut (I do not
> want the simulation to be much slower), reduce ngfft manually (anything
> else to be considered?), or in some other manner?

Since you are not optimizing the size and shape of your cell
(you set optcell=0), you do not need to use dilatmx=1.2 .
So, drop dilatmx , or set dilatmx=1.0
> 
> 2) 
> /*
>  fconv : at Broyd/MD step  24, gradients have not converged yet.
>   max grad (force/stress) = 4.3791E-03 > tolmxf= 5.0000E-05 ha/bohr
> (free atoms) 
> */
>    The grad is too large after 24 steps! Is my tolmxf too large? maybe
> result from 5.0D-3 or 5.0D-4 is convictive?

I suspect that your convergence on the forces is not well controlled,
using toldfe 1.0d-6 . You would better choose a stopping criterion
for the SCF convergence based on the forces, i.e.
toldff 5.0d-6
This will give sufficiently accurate forces to propagate the
nuclei, and to achive convergence of atomic positions
with target residual forces less than 5.0d-5 .
> 
> 3)
> /*
>  Non-SCF iterations; k pt #    2  k=  0.50000  0.50000  0.37500  band
> residuals:
>  res:  4.72E-10  3.31E-10  4.48E-10  4.66E-10  3.22E-10  4.33E-10
> 3.62E-10  5.60E-10 
> ......
> At SCF step    9, etot is converged :
>   for the second time, diff in etot=  7.844E-08 < toldfe=  1.000E-06 
> */
>     A simple question, why the iteration step is called "Non-SCF" and
> "SCF" at the same time?

The residuals that are printed correspond to the residual on the
wavefunctions in the Schrodinger equation (in a fixed potential,
so, non self-consistent).
On the other hand, there is an outer loop, that leads to
self-consistency of the potential, governed by "nstep", and different
convergence criteria (toldfe,tolvrs,toldff). Here, you choose
toldfe, and the fulfillment of the corresponding criterion is indicated.

Good continuation,
Xavier Gonze