OT produces negative gap for 1eV gapped semiconductor?

[Nicholas Winner]

Hello everyone. I have a semiconductor, BeSiP2, which has a band gap slightly above 1eV in the GGA approximation (https://materialsproject.org/materials/mp-1009085/).

I calculate a vacancy, V_{P}^{+3}, first using diagonalization. Which produces a band gap around 1eV, indicating that the charge does not delocalize, and so OT should still be appropriate. When I switch to OT, however, there is a negative homo/lumo gap and I can see it in the eigenvalues:

0.17131297      0.17154145      0.18362819      0.18519005

 Fermi Energy [eV] :    5.039278

  Lowest Eigenvalues of the unoccupied subspace spin            1

 -----------------------------------------------------

  Reached convergence in            1  iterations

       0.18499549      0.22295666      0.23850766      0.25228060

This is bizarre to me. OT is usually very reliable for me. So this is a strange inconsistency, and maybe its due to how I'm running the calculation. For OT I use the full_all preconditioner, and the rest defaults. For diagonalization I the section at the bottom. Anyone have any guesses as to why there is this discrepancy? 

-Nick

&SCF

    MAX_SCF 200

    added_mos 100

    eps_scf 1e-6

    max_diis 15

    &diagonalization

    &end

    &mixing

    &end

    &smear

        elec_temp 300

        method fermi_dirac 

    &end

&end

[Matt Watkins]

Hard to say without seeing more. I guess it has converged to the wrong electronic state somehow.

Did you restart from the Diag wfn file? You could try starting from an atomic guess instead if so and see what that produces.

That the LUMOs converge in one iteration seems suspicious somehow.

Matt

[Nicholas Winner]

Matt. I've attached the output files for my runs. I tried OT in 3 ways, first restarting from the diag wfn, second using the atomic guess (It shows a restart, but restarted from unconverged OT). The atomic guess was unable to converge, but the unconverged result shows the same negative band gap. The input for the OT run is also attached.

Thoughts?

-Nick

[Nicholas Winner]

Sorry, something isn't working with Google Groups, and the files won't send. They might come in bursts as I try.

[Nicholas Winner]

[Matt Watkins]

The MO_CUBE output form the diag cal is misleading. If you look at a pdos file you can see that there is fractional occupancy and the diag calc is smearing the last electrons over the conduction band min.

So the V3+ state seems not stable at PBE level of theory.

This is what OT shows too.

Matt

On Friday, June 4, 2021 at 3:22:04 AM UTC+1 nwi...@berkeley.edu wrote:

[Nicholas Winner]

Ah that makes sense. Thanks a lot Matt.

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/f066d01c-5ea8-43b6-a610-8b0bd7d807dfn%40googlegroups.com.