The ABINIT Users Mailing List ( CLOSED )

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

On 04 Oct 2007, at 06:38, waigen zhang wrote:

> Dear all,
> I am not sure if this problem asked  by somebody else or ...
> As far as I know, ecut and kpoint convergence test is needed before  
> any physical calculation. How about if we keep changing the  
> pressure (volume)? Say if pressure=0 GPa. ecut=20,ngkpt=4X4X4 is  
> enough. How about when pressure=100 GPa (say volume is 50% of zere  
> pressure volume)? Should we
> double check the ecut and ngkpt before any calculation?
> If we need to check the convergence when volume changes(beyond the  
> ecutsm effect),is there any good way to do that?
> Regards,
> WG.ZHANG

Dear Waigen Zhang,

First a comment ...
As a general rule, the convergence with respect to ecut and k-point  
sampling (and tsmear, in the
case of a metal) must always be checked FOR THE SPECIFIC PROPERTY in  
which one is interested.
(To make it clear : if you want to obtain some property, be it phonon  
frequencies, piezoelectric constants,
 or transitions pressure, etc , within 1%, it would be wrong to  
determine the ecut and k-points
by checking that the total energy is within some tolerance ... unlike  
sometimes advised ...)
So, independently of an answer on your e-mail, I suggest you should  
define for yourself which
property you are interested in, and understand its level of convergence.
Theorists like experimentalist should be able to deliver error bars  
on their results.

Well, with a decrease of 50% of your volume, the band structure might  
be strongly affected : a gap might
be decreased by a factor of two, or perhaps a insulator to metal  
transition will occur. Due to such effect,
the k point sampling (+tsmear) might have to be changed a lot between  
0 GPa and 100 GPa.
Of course, perhaps not ... but you should
definitely check (for the property that you are interested in !).
Concerning the ecut, since it is a property mostly related to the  
pseudopotential, it should not be affected.
However, an a posteriori check will not hurt.
In both case, if you see that the behaviour of the property of  
interest is rather smooth,  I would advice to check
convergence only at the end points of the pressure range, namely 0  
GPa and 100 GPa. On the contrary,
if you have several phases, please check the k point convergence  
(+tsmear) for each of them.

Best,
Xavier