[QE-users] Instability in HSE calculations on FeO

[Peter Stoeckl]

Dear all,

I have been attempting to calculate the band structure (and consequently
the band gap) of FeO, using HSE XC and ONCV pseudopotentials, in QE v.
6.2.1 and 6.4.1 (plus Wannier90 to interpolate the band structure).
However, the underlying SCF calculations have proven quite volatile,
frequently but inconsistently running into "dexx < 0" errors for particular
values of `nk` and `nq` – which makes a proper convergence of the results
more difficult.

I've tried a fair number of variations to circumvent the issue, with
minimal success (among them: changing the energy cutoff, convergence
threshold, smearing type, or EXX adaptive threshold – both varying initial
threshold and removing it altogether.) Occasionally, a different
`adaptive_thr` will avoid dexx < 0, but more often it makes little
difference (and sometimes does worse).

In addition, I find that changing from smearing to tetrahedral occupations
leads to noticeable qualitative differences in energy levels and resulting
band structure. While the results with tetrahedral occupation seem more
reasonable, the SCF calculations then also tend to converge slower and/or
be more prone to dexx < 0.

Is this at all expected behavior? Any suggestions to improve these results
would be greatly appreciated.

Sample Input (fails with dexx<0 for either occupation method):
 &control
    calculation = 'scf'
    restart_mode='from_scratch',
    prefix='FeO',
    pseudo_dir = '[...]',
    outdir = '[...]',
    verbosity = 'high'
 /
 &system
    ibrav=  0,
! lattice constant taken from GGA optimization
    A = 4.3100,
    nat=  4, ntyp= 3, nbnd=30,
    nspin = 2,
    starting_magnetization(1)= 0.0,
    starting_magnetization(2)= 0.5,
    starting_magnetization(3)=-0.5,
!    occupations='tetrahedra',   ! alt.
    occupations='smearing', smearing='mv', degauss=0.01,
    ecutwfc = 90.0,
    ecutrho = 360.0,
    input_dft='hse',
    exxdiv_treatment='gygi-baldereschi',
    x_gamma_extrapolation = .true.,
    nqx1=2, nqx2=2, nqx3=2
 /
 &electrons
    mixing_mode = 'plain'
    mixing_beta = 0.3
    conv_thr =  1.0d-6
    mixing_fixed_ns = 0,
    adaptive_thr = .true.
 /
! Rhombohedral 4-atom AFM cell - cf. PW example08
CELL_PARAMETERS {alat}
0.50 0.50 1.00
0.50 1.00 0.50
1.00 0.50 0.50
! SG15 ONCV pseudopotentials (scalar-rel)
ATOMIC_SPECIES
 O1   1.  O_ONCV_PBE-1.0.upf
 Fe1  1.  Fe_ONCV_PBE-1.0.upf
 Fe2  1.  Fe_ONCV_PBE-1.0.upf
ATOMIC_POSITIONS {crystal}
 O1  0.25 0.25 0.25
 O1  0.75 0.75 0.75
 Fe1 0.0  0.0  0.0
 Fe2 0.5  0.5  0.5
K_POINTS {automatic}
6 6 6 0 0 0

[End of sample input]

Peter Stoeckl
Department of Physics
University of Minnesota
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200610/0e3758b4/attachment.html>