7
$\begingroup$

What smearing would be ideal for a bcc W with interstitial O atom. I am running 3x3x3 unit cells with one N interstitial. I am having convergence issues in relax and vc-relax, but not in scf. Could this be due to smearing? I have tried mv and cold. For an additional interstitial atom, should we change or consider charge differences? There aren't many examples of interstitial atom studies other than papers which don't mentioned anything in particular about these.

 &control
    calculation = 'vc-relax',
    etot_conv_thr =   1e-05
    forc_conv_thr =   1e-04
    tprnfor = .true.
    tstress = .true.
    verbosity = 'high'
    pseudo_dir = '.',
    !prefix='w',
 /
 &system
    ibrav = 1,
    celldm(1) = 18.000,
    nat = 55,
    ntyp= 2,
    ecutwfc = 50.0, ecutrho = 500.0,
    occupations='smearing', smearing='mv', degauss=0.01,
    !nbnd = 8,
    !nspin=2,
    starting_magnetization(1)= 0.6
 /
 &electrons
    mixing_beta = 0.7
    conv_thr=1e-6
 /
 &ions
    ion_dynamics='bfgs'
 /
 &cell
    cell_dynamics='bfgs'
    cell_dofree='volume'
 /
ATOMIC_SPECIES
 W 183.84 W.pbe-spn-rrkjus_psl.1.0.0.UPF
 O 15.999 O.pbe-n-rrkjus_psl.1.0.0.UPF
ATOMIC_POSITIONS crystal
W      0.000000   0.000000   0.000000
W      0.166667   0.166667   0.166667
W      0.000000   0.000000   0.333333
W      0.166667   0.166667   0.500000
W      0.000000   0.000000   0.666667
W      0.166667   0.166667   0.833333
W      0.000000   0.333333   0.000000
W      0.166667   0.500000   0.166667
W      0.000000   0.333333   0.333333
W      0.166667   0.500000   0.500000
W      0.000000   0.333333   0.666667
W      0.166667   0.500000   0.833333
W      0.000000   0.666667   0.000000
W      0.166667   0.833333   0.166667
W      0.000000   0.666667   0.333333
W      0.166667   0.833333   0.500000
W      0.000000   0.666667   0.666667
W      0.166667   0.833333   0.833333
W      0.333333   0.000000   0.000000
W      0.500000   0.166667   0.166667
W      0.333333   0.000000   0.333333
W      0.500000   0.166667   0.500000
W      0.333333   0.000000   0.666667
W      0.500000   0.166667   0.833333
W      0.333333   0.333333   0.000000
W      0.500000   0.500000   0.166667
W      0.333333   0.333333   0.333333
W      0.500000   0.500000   0.500000
W      0.333333   0.333333   0.666667
W      0.500000   0.500000   0.833333
W      0.333333   0.666667   0.000000
W      0.500000   0.833333   0.166667
W      0.333333   0.666667   0.333333
W      0.500000   0.833333   0.500000
W      0.333333   0.666667   0.666667
W      0.500000   0.833333   0.833333
W      0.666667   0.000000   0.000000
W      0.833333   0.166667   0.166667
W      0.666667   0.000000   0.333333
W      0.833333   0.166667   0.500000
W      0.666667   0.000000   0.666667
W      0.833333   0.166667   0.833333
W      0.666667   0.333333   0.000000
O      0.416666   0.500000   0.500000
W      0.833333   0.500000   0.166667
W      0.666667   0.333333   0.333333
W      0.833333   0.500000   0.500000
W      0.666667   0.333333   0.666667
W      0.833333   0.500000   0.833333
W      0.666667   0.666667   0.000000
W      0.833333   0.833333   0.166667
W      0.666667   0.666667   0.333333
W      0.833333   0.833333   0.500000
W      0.666667   0.666667   0.666667
W      0.833333   0.833333   0.833333
! this is a comment that the code will ignore
K_POINTS automatic
4 4 4 1 1 1
$\endgroup$
16
  • 2
    $\begingroup$ I don't know much about QE, but W psuedopotentials tend to be difficult (in my experience). You may need more valence electrons to stabilize your calculation maybe? $\endgroup$ Jan 2, 2021 at 19:23
  • 1
    $\begingroup$ For studying effects due to an interstitial atom, I'm not sure as to why you've chosen a vc-relax calculation. Usually, the unit cell of the unaltered crystal will be subjected to vc-relax and with the optimized unit-cell parameters you would construct the supercell. Include the interstitial atom and then do a "relax calculation". Atleast this is how I've seen it being done. $\endgroup$ Jan 2, 2021 at 20:13
  • 1
    $\begingroup$ Since youre having convergence issues, could you try to change the default diagonalization form "david" to "cg" using diagonalization = "cg" in your &electrons namelist. It has helped me sometimes atleast. $\endgroup$ Jan 2, 2021 at 20:17
  • 2
    $\begingroup$ @Dr.Viper I think his main point was that you should relax the pure cell before addition, but also keep it fixed after addition. This assumes you want to model the dopant in its non-interacting dilute limit rather than relaxing the crystal to account for the dopant being incorporated uniformly. Think of surface calculations where the cell dimensions do not change, even after absorption of an adsorbate if this helps you think about it. The adsorbate in high coverage may play a slight role in the surface cell, but in the dilute limit it should definitely not influence things. $\endgroup$ Jan 2, 2021 at 22:54
  • 1
    $\begingroup$ @Dr.Viper Scratch my suggestion, you already are using a spn potential which means s and p states are involved as valence. $\endgroup$ Jan 3, 2021 at 0:11

1 Answer 1

3
$\begingroup$

There are many reasons behind not converging. Sometimes structure doesn't converge if you describe them wrong. So, in my opinion, the best way to converge a structure is to study it clearly. Moreover, this convergence also varies for observing different properties too. One common trick is to check smearing in your system. For your system check if it converges for "gaussian" or any other smearing. Another reason for not converging is the lack of an appropriate k point mesh grid for your system. Increasing the K point mesh grid may improve the convergence in less iteration though it consumes more computing time. You need to be careful choosing the k point mesh grid if it supports your system. Another thing you can do is to increase iteration from default. Also, you can change the pseudopotential type.

$\endgroup$

You must log in to answer this question.

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