Page 1 of 1

Hybrid problem

Posted: Sat Dec 05, 2015 4:43 pm
by gak
Hey,

I am having some problems trying to run PBE0 and HSE06 simulations of silicon. I am getting lattice parameters and a band gap that is significantly different to literature. I have calculated the lattice parameter by a geometry relaxation and also by energy minimising various lattice parameters, the result is basically the same. Here are my relaxation inputs;

INCAR :

ISMEAR = 0
SIGMA = 0.01
GGA = PE
LHFCALC = .TRUE. ;
HFSCREEN = 0.2
ALGO = ALL # I have tested both ALL and DAMPED, both give the same answer
TIME = 0.3 ;
ENCUT = 640
NBANDS = 32 # I increased my bands as there was one paper that quoted using a
NKRED = 21
EDIFF = 1E-06
NELMIN = 10
NSW = 50
IBRION = 2
ISIF = 3
PRECFOCK = Accurate



KPOINTS;
K-Points
0
Monkhorst Pack
21 21 21
0 0 0


POSCAR;
system Si
5.43
0.5 0.5 0.0
0.0 0.5 0.5
0.5 0.0 0.5
2
cart
0.00 0.00 0.00
0.25 0.25 0.25


This gives me a lattice parameter of 5.37A for both PBE0 and HSE, my band gaps are also about twice as big as literature results.

I am running a 60 core simulation (on 3 nodes, each with 24 cores). I also tried to run these using NPAR / KPAR, and this gave me a completely different set of results (I saw on this forum that the NPAR shouldn't be used in conjunction with hybrid's though). I also tried using a 24x24x24 gamma centre k-point grid (with a 24x24x24 q-point grid), but this was also wrong.

Is there something obvious that I'm missing? Any help is much appreciated.

Re: Hybrid problem

Posted: Mon Dec 07, 2015 8:43 pm
by gak
Here are some of my other results, which are significantly different to the ones above (I'm using the same 60 processors described above),


HSE06 ;
"
ISMEAR = 0
SIGMA = 0.01
GGA = PE
LHFCALC = .TRUE. ; HFSCREEN = 0.2 ;
ALGO = ALL ; TIME = 0.3 ;
ENCUT = 640
NBANDS = 16
NPAR = 8
KPAR = 12
NKRED = 24
EDIFF = 1E-06
EDIFFG = -0.02
NELM = 200
"

Here I'm using a gamma centred 24x24x24 grid, I get a lattice parameter of 5.417 A (which is in good agreement to a literature value I found of 5.419 A), and a band gap of over 2.8 eV.



PBE0;
"
ISMEAR = 0
SIGMA = 0.01
GGA = PE
LHFCALC = .TRUE. ;
ALGO = ALL ; TIME = 0.3 ;
ENCUT = 500
NBANDS = 16
NPAR = 8
KPAR = 12
NKRED = 24
EDIFF = 1E-06
EDIFFG = -0.02
NELM = 200
"

This gives a lattice parameter of 5.418A (the same literature as the above HSE06 result quotes 5.433A), and a band gap of 3.6 eV. So my band gaps in these cases are more than twice the expected result. I really can't think of anything else here, especially as these are given as a tutorial example on the website.

Re: Hybrid problem

Posted: Wed Dec 09, 2015 3:53 pm
by kelum
Hi,
you are using NKRED equal to your number of kpoints in each direction. That means that for the hybrid part you are using one or few k-points.
This makes the hybrid contribution inaccurate. For 24^3 kpts and HSE, reasonable values for NKRED are most likely 3 or 4, possibly 6, so that you'd use
a 8x8x8 k-points (or 6x6x6 or 4x4x4 k-points) for the hybrid part. As written in the paper by Paier (2006 JCP), for PBE0 NKRED is not much recommended.
Best,
jik

Re: Hybrid problem

Posted: Fri Dec 18, 2015 2:39 pm
by gak
kelum wrote:Hi,
you are using NKRED equal to your number of kpoints in each direction. That means that for the hybrid part you are using one or few k-points.
This makes the hybrid contribution inaccurate. For 24^3 kpts and HSE, reasonable values for NKRED are most likely 3 or 4, possibly 6, so that you'd use
a 8x8x8 k-points (or 6x6x6 or 4x4x4 k-points) for the hybrid part. As written in the paper by Paier (2006 JCP), for PBE0 NKRED is not much recommended.
Best,
jik

Hey,

Thank you very much, that has sorted my (silly) mistake. One more question. I am trying to trying to do some calculations on silicon at elevated temperatures (I know DFT is a groundstate theory), so I have applied an electronic temperature using Fermi-Dirac smearing. It is very high, 2 eV. So I made sure I had empty bands (I need 25+), but when I retested the convergence of my ENCUT, it starts to oscillate after a certain point for both PBE0 and HSE, here is an example (numbers on the right are ENCUT);

free energy TOTEN = -17.96385021 eV | 350
free energy TOTEN = -17.96526757 eV | 375
free energy TOTEN = -17.96616419 eV | 400
free energy TOTEN = -17.96658290 eV | 425
free energy TOTEN = -17.96672994 eV | 450
free energy TOTEN = -17.96673496 eV | 475
free energy TOTEN = -17.96669010 eV | 500
free energy TOTEN = -17.96667417 eV | 525
free energy TOTEN = -17.96665050 eV | 550
free energy TOTEN = -17.96666595 eV | 575
free energy TOTEN = -17.96668197 eV | 600
free energy TOTEN = -17.96669841 eV | 625
free energy TOTEN = -17.96671119 eV | 650
free energy TOTEN = -17.96671882 eV | 675
free energy TOTEN = -17.96671965 eV | 700


I thought wave-function cut-off was variational, is this something that could be expected for non-equilibrium configurations on hybrids? Thanks.