Dear Michal
There are a few tricks to run EXX simulations in QE
First of all, in my experience best results are obtained by using
norm-conserving pseudopotentials. I've not tested US or PAW
implementations recently, but they used to be unstable and slower than
NC, even if NC requires much higher ecutwfc.
Then
ecutwfc=90.0, (<-- this is the converged value for *my* pseudopotentials...)
ecutfock=90.0,
ecutfock is the cutoff for the Fock operator, it *strongly* affects
the calculation time&resources. As a rule of thumb ecutfock=ecutwfc is
a sensible choice for geometry optimizations and
ecutfock=1.5~2*ecutwfc is a sensible convergence range for accurate
electronic properties. The default value (ecutfock=ecutrho) is a huge
loss of time and resources!
adaptive_thr=.true.
ensures a significant speedup for EXX loops.
I used this setup for the geometry optimization of a cyclic dipeptide
(27 atoms, 78 electrons) on a 4-cores/16 GB RAM system with QE 6.6
&control
calculation = 'relax'
restart_mode='from_scratch',
prefix='$FILE',
pseudo_dir = '$PSEUDO_DIR/',
outdir='$TMP_DIR/',
tprnfor=.true.,
nstep=200,
disk_io='low',
wf_collect=.true.,
max_seconds=40000,
/
&system
ibrav=1, celldm(1)=40.0000,
nat=27 ntyp=4,
ecutwfc=90.0,
ecutfock=90.0,
occupations='smearing', degauss=0.01, <-- comment it out if you
don't need unoccupied orbitals
nspin=1,
input_dft='b3lyp'
vdw_corr='grimme-d3',
assume_isolated='mp', <-- useful for an estimate of the vacuum level
/
&electrons
diagonalization='david',
mixing_mode='plain',
mixing_beta=0.1,
conv_thr=1.0d-6,
electron_maxstep=100
scf_must_converge=.false., <-- when convergence of the first GGA
step is very difficult, use with care!
adaptive_thr=.true.
/
&ions
ion_dynamics='bfgs'
/
ATOMIC_SPECIES
O 15.999 O.blyp-mt.UPF
N 14.007 N.blyp-mt.UPF
C 12.011 C.blyp-mt.UPF
H 1.008 H.blyp-vbc.UPF
ATOMIC_POSITIONS {angstrom}
...
K_POINTS {gamma}
The output should print the expected memory usage (it is often
slightly overestimated)
Estimated total dynamical RAM > 7.35 GB
As you can see, such calculation requires up to 8 GB RAM, and a little
more if you increase ecutfock to 135 Ry (i.e., 1.5*ecutwfc). Keep in
mind that QE is not designed for calculations of isolated systems
surrounded by a lot of vacuum space, and it uses for small molecules
many more resources than a GTO code. Nevertheless, I've satisfactorily
performed QE simulations of a lot of small and large molecular systems
using EXX functionals such as B3LYP or HSE...
HTH
Giuseppe
Quoting Husak Michal <[email protected]>:
I would like to ask what is the typical time difference between
running PBE and
and the hybrid PBE0 ...
Using CASTEP the PBE0 takes typically 200-1000 x more time than PBE ...
This is OK , I run CASTEP on a 1000 node supercomputer and work is
done in a few
days typically ...
I had run PBE and PBE0 scf on super simple system (Al generated by
Material Cloud www, diamond generated by me) ...
I run in MP mode on 4 core CPU system ...
PBE - 7.5 second
PBE0 - run 24 hours and not yet finished
PBE0 on the target molecule (alanine) crashed immediately with
segmentation failure ...
Do I something wrong, is there some magic in setting hybrid
functional or do i relay need a supercomputer ? I believed I can
just add PBE0 request to tested PBE setup ...
Can anybody send me a working sample PBE0 calculation setup for
small organic molecular crystal ?
Michal Husak
UCT Prague
_______________________________________________
Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
users mailing list [email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users
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: <[email protected]>
_______________________________________________
Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
users mailing list [email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users
