[Pw_forum] cutoff convergence of sapphire for elastic properties

[Axel Kohlmeyer]

On Tue, May 8, 2012 at 12:42 PM, Jörg Buchwald
<joerg.buchwald at iom-leipzig.de> wrote:
>> Considering the normal errors in DFT, it's not worth trying to
>> converge the stress to the 1 kbar accuracy you're trying to achieve.
>
> But to get the elastic constants in the elastic regime, i would
> like then apply strains of serveral per mill, which is of the same
> order of magnitude, i.e. also corresponds to changes of the stress
> tensor of the order of 1 kbar, which means that errors in the kbar
> range would be too high.

question is: do you have the same absolute
error of the same magnitude and sign on all
stress tensor computations, so that the
*relative* error would be much smaller?

remember, that when you do e.g. a geometry optimization
you use forces (derivatives of the potential energy) that
are converged long before your energy is converged, so
by your argument, you would have to use a much higher
cutoff for that as well.

> An alternative could be the calculation of the elastic constants using
> the second derivative of the energy. But this won't work for big
> supercells due to the computation time and the number of measuring
> points needed.

see the comment about slow and fast converging
properties. you probably don't need to converge the
energy as tightly for this due to error cancellation, too.

however, convergence and error cancellation is not
always easily predictable. some careful experimentation
is often better than using either brute force, guessing
or faith in some random person's ramblings. ;-)

cheers,
     axel.


>
>
>>
>> On Mon, May 7, 2012 at 10:07 AM, Nicola
>> Marzari<nicola.marzari at epfl.ch> wrote:
>>
>> >
>> > How much does 1 kbar error translates into an error in lattice
>> > parameter? (keep atoms fixed, using relative coordinates, cutoff
>> > fixed, and expand celldm(1) by 0.3% - what's the change in stress?
>> > that change should be very well converged)
>> >
>> > --
>> Michael Mehl
>> US Naval Research Laboratory
>> Washington DC
>> (Home email)
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum



-- 
Dr. Axel Kohlmeyer
akohlmey at gmail.com  http://goo.gl/1wk0

College of Science and Technology
Temple University, Philadelphia PA, USA.