[QE-users] HSE OPTIMIZATION

[Giuseppe Mattioli]

Dear Ashley Cooper (remember that we like to know who you are and  
where you work...)

There is an old README file in this directory

Your_path_to_QE/7.4.1/PW/examples/EXX_example

In this file, that may be not updated since a lot of time, thus  
possibly wrong, it is written that:

  WHAT PROPERTIES CAN I COMPUTE ?
   Energy and forces (thanks to Hellmann-Feynman theorem forces do not
   require extra calculations). In principle also stresses but the
   corresponding formulas have not yet been coded.
   So structural optimization is OK if the cell shape is kept fixed.

Provided that the README is right, you can either fit by hand some  
energy vs volume curve performing a series of relax calculations at  
different lattice parameters to find the minimum, or you can optimize  
your lattice with vc-relax using your favorite GGA functional, and  
then relax the obtained structure using HSE. I would do the former in  
the case of cubic cells having a single lattice parameter, and the  
latter in the case of hexagonal, tetragonal, orthorombic, ...

This said, you can save significant computational resources using

1) ONCV pseudopotentials (look around for the sg15 database) which  
should require ecutwfc significantly lower than 140 Ry (it is useless  
to specify ecutrho in the case of NC pseudopotentials).

2) ecutfock, a specific cutoff for Fock integrals. You can obtain good  
results using ecutfock = 1~2*ecutwfc, with ecutfock closer to  
1*ecutwfc that is good enough for geometry optimization and ecutfock  
closer to 2*ecutwfc that is good enough for property calculations.

3) ibrav  =  0. Don't use it. Exploit lattice symmetry using a  
meaningful ibrav = 4 label for hexagonal cells.

HTH
Giuseppe


Quoting Barsha Pal <barsha.pal at aus.ac.in>:

> I am trying to optimize an hexagonal unit cell with HSE06 functional. But
> the optimization is running for 12 days and still stuck at bfgs step 2,
> with scf step count 34, the optimization taking a long time. The input
> specifications are:
>
> &CONTROL
>     calculation   = "vc-relax"
>     restart_mode  = "from_scratch"
>     forc_conv_thr =  1.00000e-04
>     etot_conv_thr =  1.00000e-05
>     outdir  = "/"
>     prefix        = "hse_opt"
>     pseudo_dir    = "/"
>     verbosity     = "high"
> /
>
> &SYSTEM
>     celldm(1)   =  15.36942147
>     degauss     =  1.00000e-02
>     ecutrho     =  560
>     ecutwfc     =  140
>     ibrav       =  0
>     nat         =  12
>     ntyp        =  3
>     occupations = "smearing"
>     smearing    = "gaussian"
>     input_dft   = "hse"
>     exx_fraction = 0.25
>     screening_parameter = 0.106
>     exxdiv_treatment =  "gygi-baldereschi"
>     x_gamma_extrapolation = .true.
>     nqx1 = 2,
>     nqx2 = 2,
>     nqx3 = 1,
>     assume_isolated = "2D"
> /
>
> &ELECTRONS
>    conv_thr         =  1.00000e-08
>    electron_maxstep =  300
>    mixing_beta      =  7.00000e-01
>    diagonalization  ='david'
>    mixing_mode    = 'plain'
> /
>
> &IONS
>     ion_dynamics = "bfgs"
> /
>
> &CELL
>     cell_dofree    = "2Dxy"
>     cell_dynamics  = "bfgs"
>     press          =  0.00000e+00
>     press_conv_thr =  0.5
> /
>
> K_POINTS {automatic}
>   12  12  1  0 0 0
>
> ATOMIC_SPECIES
> A     1   A.pbe-mt_fhi.UPF
> B     1   B.pbe-mt_fhi.UPF
> C     1   C.pbe-mt_fhi.UPF
>
>
>
>
> Previously I tried to run with q points 3 3 1 and the optimization was
> running at bfgs step 2 for 1 month,
>
> How can I use the functional and reduce the computational time? Please let
> me know.
>
>
>
>
> Ashley Cooper
> Phd



GIUSEPPE MATTIOLI
CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
Via Salaria Km 29,300 - C.P. 10
I-00015 - Monterotondo Scalo (RM)
Mob (*preferred*) +39 373 7305625
Tel + 39 06 90672342 - Fax +39 06 90672316
E-mail: <giuseppe.mattioli at ism.cnr.it>