5
$\begingroup$

At the moment I'm performing various convergence tests of a cation-solvent-anion complex in ASE using the GPAW calculator, and found strange results for energy vs. vacuum padding as shown in the figure below. I consider a parameter converged when ΔE < 1 meV/atom.

The isolated Na atom converges, but acetonitrile jumps "randomly", hexafluorophosphate shows an artifact pattern of some sort, and the complex looks like a combination of the three.

I have checked that the force criterion of the ionic loop is converged, and tried:

  • breaking up the complex

  • appling ExtraVacuumPoissonSolver (no difference in energy with or without)

  • simplifying the script. I don't do more then read and write, and:

    opt.run(fmax = args.fmax, steps=args.maxsteps)
E_gas = mol.get_potential_energy()

enter image description here

For context, here is the convergence tests of the electronic loop criterion, the ionic loop criterion, and vacuum padding for four Na-solvent-hexafluorophosphate complexes (a nitrogen, sulphur, oxygen-, and fluorine-containing species (methyl 2-cyanoacetate, 1-ethenylsulfonylethane, 2,6-dimethoxycyclohexa-2,5-diene-1,4-dione, and 2-(2-ethoxyethoxy)-1,1-difluoroethane, respectively)).

enter image description here

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ which basis set are you using? $\endgroup$
    – franz
    Commented Oct 29 at 8:39

1 Answer 1

Reset to default
5
$\begingroup$

The issue was that I used too large of a grid spacing at 0.26 Å. This value I got from a utility tool from GPAW, cutoff2gridspacing. From an old convergence test with VASP on similar systems the cutoff energy converged at 550 eV. Plugging 550 eV into cutoff2gridspacing yielded 0.26 Å. I have now redone a convergence test of the grid spacing, resulting in a much lower value of 0.13 Å.

Shortcuts are never the answer..

Improve this answer
$\endgroup$
5
  • $\begingroup$ Remember that the cut-off energy is directly related to the pseudopotential you use, and is definitely not transferable between different pseudopotentials. For real space codes, it is also affected by the order of spline/basis function fit. $\endgroup$
    – Phil Hasnip
    Commented Nov 25 at 15:32
  • $\begingroup$ Can you point me to where I can find the order of spline/basis function fit in the GPAW documentation? $\endgroup$
    – awwnoh
    Commented Nov 26 at 14:18
  • $\begingroup$ Are you using a real-space basis? For GPAW it depends which mode you're using (gpaw.readthedocs.io/documentation/basic.html). $\endgroup$
    – Phil Hasnip
    Commented Nov 27 at 1:35
  • $\begingroup$ I'm using fd mode, so as I understand it, the basis set is only used to initialise the wave functions. By "order", do you mean whether I use "sz", "dz", "dzp", and so on? $\endgroup$
    – awwnoh
    Commented Nov 29 at 8:31
  • $\begingroup$ fd mode is a different real-space basis. I haven't managed to find the exact stencil used in the documentation, but the interpolation in fd mode is with a quintic so that is not amazingly good, but not terrible either. $\endgroup$
    – Phil Hasnip
    Commented Nov 30 at 0:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .