[QE-users] Problem of HSE scf calculation with QE code

Lorenzo Paulatto paulatz at gmail.com
Fri Jul 10 11:58:43 CEST 2020


Well, you could start by using the unit cell instead of the simple-cubic 
super-cell, unless you have some specific reason for this.

In the next version of QE there will be a generalized implementation of 
exchange using localized wavefunction in real-space, although in my 
experience it is not much faster than the normal algorithm when studying 
such a small system.

If you are running on a large number of CPUs, activating bands 
parallelization (-nbnd XX) can make it much more efficient

cheers

On 7/10/20 11:36 AM, ygxu17 wrote:
> Hello Lorenzo,
> Thanks for you advice. I tried nqx1=nqx2=nqx3=2 and got the value of 
> 1.33 eV for the gamma point gap. It seems the problem comes from the 
> sampling mesh grid.
> However, as is known, HSE calculation is expensive. In my calculation of 
> TiO2, 240 atoms and gamma_only kpoints are used. How should I set the 
> nqx for this calculation from your experience?('1 1 1' fails to 
> reproduce the bandgap.) By the way, is there any advice to speed up the 
> HSE scf calculation?
> &CONTROL
>                         title = 'TiO2'
>                   calculation = 'scf'
>                  restart_mode = 'from_scratch'
>                        outdir = './1/'
>                    pseudo_dir = './'
>                        prefix = 'C'
>                       tstress = .true.
>                       tprnfor = .true.
>                     wf_collect=.true.
>                     disk_io = 'low'
>                      verbosity = 'high'
>   /
>   &SYSTEM
>                         ibrav = 0
>                           nat = 240
>                          ntyp = 2
>                       ecutwfc = 90
>    ecutfock = 180
>                         nosym = .true.
>                         tot_charge = 0
> !nbnd=20
> occupations='smearing', smearing='gaussian', degauss=0.002
>      input_dft = 'hse'
>           nqx1 = 1
>           nqx2 = 1
>           nqx3 = 1
>     exx_fraction = 0.25
>   /
>   &ELECTRONS
>               diagonalization = 'david'
>                  conv_thr = 1.0D-6
>    adaptive_thr = .true.
>          mixing_beta = 0.7
>    electron_maxstep=100
>   /
> &IONS
>   ion_dynamics = 'bfgs'
> /
> ATOMIC_SPECIES
> Ti 47.867 Ti.SG15.PBE.UPF
> O 15.999 O.SG15.PBE.UPF
> 
> K_POINTS gamma
> 
>   CELL_PARAMETERS bohr
>     22.2322500562796       0.000000000000000E+000  0.000000000000000E+000
>    0.000000000000000E+000   24.3827575075721       0.000000000000000E+000
>    0.000000000000000E+000  0.000000000000000E+000   57.4954854151373
> 
>   ATOMIC_POSITIONS bohr
>   Ti   0.000000000000000E+000   6.09568937689302        2.41706800158272
>   Ti   0.000000000000000E+000  0.000000000000000E+000   7.80624023418306
>   Ti   0.000000000000000E+000   6.09568937689302        14.8977770162714
>   Ti   0.000000000000000E+000  0.000000000000000E+000   20.6787177547025
>   Ti   0.000000000000000E+000   6.09568937689302        27.2633365743639
>   Ti    2.77903125703495       0.000000000000000E+000   1.45370321265824
> ...
> 
> Best regards,
> Greg Xu
> 
> On 7/10/2020 16:17,Lorenzo Paulatto<paulatz at gmail.com> 
> <mailto:paulatz at gmail.com> wrote:
> 
>                   nqx1 = 1
>                   nqx2 = 1
>                   nqx3 = 1
> 
>     Hello Greg,
>     these choce of parameters is cheap, but not very well converged. Did
>     you
>     try to use a finer grid for exchange?
> 
>     cheers
> 
> 
>     -- 
>     Lorenzo Paulatto - Paris
>     _______________________________________________
>     Quantum ESPRESSO is supported by MaX
>     (www.max-centre.eu/quantum-espresso)
>     users mailing list users at lists.quantum-espresso.org
>     https://lists.quantum-espresso.org/mailman/listinfo/users
> 
> 
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
> 

-- 
Lorenzo Paulatto - Paris


More information about the users mailing list