[Pw_forum] Hybrid DFT calculations in QE v 6.2 and PAW datasets
Paolo Giannozzi
p.giannozzi at gmail.com
Mon Feb 5 08:32:05 CET 2018
PAW with hybrid functionals seem to work (or at least, I have no clear
evidence they don't) but they are less tested than USPP and even less
tested than NC. Both PAW and USPP tend to give the infamous "dexx is
negative" error more often than not, for unclear reasons (the computed
quantity should be positive definite). One can turn the check off, take the
absolute value for dexx and continue, typically to convergence, but no
warranty. Note that currently both USPP and PAW are quite slow (less so
with option "tqr") if used with hybrid functionals. If you use them with a
high cutoff (80 Ry is reasonable for norm-conserving H,C,N,O, very high for
PAW), it is much faster to use NC, unless you have some compelling reason
to use USPP or PAW
Paolo
On Fri, Jan 26, 2018 at 1:06 AM, Luc LeBlanc <Luc.LeBlanc at dal.ca> wrote:
> Hello,
>
> I have recently been attempting to perform calculations within Quantum
> ESPRESSO v 6.2 using hybrid PBE0 DFT on a couple of test systems. However,
> I am encountering the following problem. For the following test case, a
> diamond crystal structure (input can be found below), I have had the
> following error show up in the output:
>
> "dexx is negative! Check that exxdiv_treatment is appropriate for the
> system, or ecutfock may be too low"
>
> Now, I realize that this error has been reported many times in previous
> years and is due to the divergence of the exact exchange operator at small
> values of q points. As I understand it, the current treatment for this
> diverging term (gygi-baldereschi) is applicable to cells that are ~cubic;
> but, the diamond crystal is cubic. Suggested workarounds to resolve the
> error above have been to change this scheme (e.g., truncate the Coulomb
> potential or use vcut_ws, increase ecutfock, shift the k-point mesh away
> from the gamma point, etc.). I have exhausted all of these possibilites,
> none of which have worked for the diamond crystal or many other crystal
> systems that I have tried. For instance, the same error occurs for the urea
> crystal structure, but not for a single urea molecule in a large vaccuum
> (both inputs can be found below) when performing SCF calculations.
>
> Because of this, I have become suspicious of the pseudopotentials that I
> am using to perform these calculations, i.e., the PBE PAW datasets (e.g.,
> the ones found on the QE website). If I instead use a PBE norm-conserving
> pseudopotential, the SCF does converge.
>
> Are there any considerations to take into account when generating PAW
> datasets as opposed to norm-conserving pseudopotentials that would cause
> this issue? Or is there something in the QE implementation itself which
> would cause this divergence in exact exchange when PAW datasets are used?
>
> If anyone has insight on the matter, help would be appreciated.
>
> Thank you,
> Luc LeBlanc
> luc_DOT_leblanc_AT_ dal_DOT_ca
> Department of Chemistry
> Dalhousie University, Canada
>
> The input file for the diamond crystal structure:
>
> &control
> title='diamond',
> prefix='diamond',
> pseudo_dir='.',
> /
> &system
> ibrav=0,
> celldm(1)=1.0,
> nat= 8,
> ntyp= 1,
> ecutwfc=80.0,
> ecutrho=800.0,
> input_dft='PBE0',
> exx_fraction=0.25,
> /
> &electrons
> conv_thr = 1d-8,
> /
> &ions
> /
> &cell
> /
> ATOMIC_SPECIES
> c 12.010700 c.UPF
>
> ATOMIC_POSITIONS crystal
> c 0.12500000 0.12500000 0.12500000
> c 0.12500000 0.62500000 0.62500000
> c 0.62500000 0.12500000 0.62500000
> c 0.62500000 0.62500000 0.12500000
> c 0.87500000 0.87500000 0.87500000
> c 0.87500000 0.37500000 0.37500000
> c 0.37500000 0.87500000 0.37500000
> c 0.37500000 0.37500000 0.87500000
>
> K_POINTS automatic
> 4 4 4 1 1 1
>
> CELL_PARAMETERS bohr
> 6.740086198163 0.000000000000 0.000000000000
> 0.000000000000 6.740086198163 0.000000000000
> 0.000000000000 0.000000000000 6.740086198163
>
> The inputs for the urea crystal and single urea molecule in a large
> vaccuum:
>
> &control
> title='urea',
> prefix='urea',
> pseudo_dir='.',
> /
> &system
> ibrav=0,
> celldm(1)=1.0,
> nat= 16,
> ntyp= 4,
> ecutwfc=80,
> ecutrho=800,
> input_dft='PBE0',
> exx_fraction=0.25,
> /
> &electrons
> conv_thr = 1d-8,
> /
> &ions
> /
> &cell
> /
> ATOMIC_SPECIES
> h 1.007940 h.UPF
> c 12.010700 c.UPF
> n 14.006700 n.UPF
> o 15.999400 o.UPF
>
> ATOMIC_POSITIONS crystal
> c 1.00000000 0.50000000 0.32747528
> c 0.50000000 0.00000000 0.67252472
> h 0.26121365 0.76121365 0.28250193
> h 0.76121365 0.73878635 0.71749807
> h 0.23878635 0.26121365 0.71749807
> h 0.73878635 0.23878635 0.28250193
> h 0.14516494 0.64516494 0.95967949
> h 0.64516494 0.85483506 0.04032051
> h 0.35483506 0.14516494 0.04032051
> h 0.85483506 0.35483506 0.95967949
> o 1.00000000 0.50000000 0.60019309
> o 0.50000000 0.00000000 0.39980691
> n 0.14704706 0.64704706 0.17768392
> n 0.64704706 0.85295294 0.82231608
> n 0.35295294 0.14704706 0.82231608
> n 0.85295294 0.35295294 0.17768392
>
> K_POINTS automatic
> 4 4 4 1 1 1
>
> CELL_PARAMETERS bohr
> 10.473277845000 0.000000000000 0.000000000000
> 0.000000000000 10.473277845000 0.000000000000
> 0.000000000000 0.000000000000 8.826950084000
>
> ####
>
> &control
> title='urea_mol',
> prefix='urea_mol',
> pseudo_dir='.',
> /
> &system
> ibrav=0,
> celldm(1)=1.0,
> nat= 8,
> ntyp= 4,
> ecutwfc=80,
> ecutrho=800,
> input_dft='PBE0',
> exx_fraction=0.25,
> /
> &electrons
> conv_thr = 1d-8,
> /
> &ions
> /
> &cell
> /
> ATOMIC_SPECIES
> h 1.007940 h.UPF
> c 12.010700 c.UPF
> n 14.006700 n.UPF
> o 15.999400 o.UPF
>
> ATOMIC_POSITIONS crystal
> c 0.00462173 0.01313468 0.00001026
> o 0.03762823 0.07042739 0.99972194
> n 0.93159137 0.00837905 0.98441030
> h 0.90253872 0.96768568 0.00623664
> h 0.90634733 0.05670150 0.98288178
> n 0.03721573 0.94767996 0.01598762
> h 0.01654374 0.90200798 0.99445500
> h 0.09169167 0.95012689 0.01719219
>
> K_POINTS automatic
> 1 1 1 1 1 1
>
> CELL_PARAMETERS bohr
> 35.000000000000 0.000000000000 0.000000000000
> 0.000000000000 35.000000000000 0.000000000000
> 0.000000000000 0.000000000000 35.000000000000
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>
--
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20180205/9e4c761f/attachment.html>
More information about the users
mailing list