Bulk potential energy by different methods

Matt W mattwa... at gmail.com
Fri Jun 2 07:50:17 UTC 2017


individual total energy are almost always meaningless, especially when 
comparing different levels of theory. But, relative energies that 
correspond to physical quantities should give the same answer, if the level 
of theory is adequate.

Your input looks OK to me, except EPS_SCF 1.0E-4 is maybe too large, check 
that the energy is well converged.

Is your density of vacancies the same in each case? Did you relax the 
structures? Is LDA a good enough density functional. Is a DZVP basis set 
good enough?


On Thursday, June 1, 2017 at 4:31:36 PM UTC+1, Sebastian Hütter wrote:
> Hi all,
> this may be a simple question, but somehow I feel like I am missing 
> something very basic here.
> I'm simply comparing bulk energies of metals (here: Al) computed over 
> XYZ-periodic unit cells obtained via DFT in cp2k and several EAM potentials 
> in LAMMPS. Different published potentials in LAMMPS all agree on an energy 
> of somewhere around -14 eV/unit cell. However, when I run the same 
> computation in cp2k (input file attached), I find energies (printed as the 
> "FORCE_EVAL ( QS ) energy") of -8.3 Ha/unit cell, or -226 eV/unit cell. As 
> this is more than one order of magnitude off, something can't be right here.
> To see if the relative energies are usable, I then tried to calculate 
> vacancy formation energy by comparing the supercell energies with and 
> without vacancies (cf. 
> https://icme.hpc.msstate.edu/mediawiki/index.php/LAMMPS_Vacancy_Formation_Energy) 
> , but that is wrong as well, while the LAMMPS solution agrees reasonably 
> well with experimental data.
> It appears as if every energy value calculated is too large by a factor of 
> 16? Why would that be the case?
> Thank your for your help,
> Sebastian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170602/5ee6936b/attachment.htm>

More information about the CP2K-user mailing list