Could you please post your V pseudo?
The mesh cutoff doesn't vary the scf, it only controls the rippling of
energy/forces.
This rippling is due to the FFT aliasing of those magnitudes
(orbs^2,neutral atom potential, core
charge for non linear core corrections) which are projected into the
grid. As the
aliasing is constant (ie during the scf the atoms don't move with
respect to
the grid) the mesh cutoff shouldn't play any role.
Best regards,
Eduardo
On 04/11/2008, at 18:05, Joachim Fürst wrote:
Hi,
Thank you for your advices.
I have also cut the system down to only a 100 atoms or so, without
any improvement. I have a pretty good idea about V position, what is
less known to me is the distortion of the sheet. I have done runs
with 220 Ry mesh, but no change. I will try very high temperature as
a final try.
Thansk again.
Joachim
-----Original Message-----
From: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta
[mailto:[EMAIL PROTECTED] On Behalf Of Marcos Verissimo Alves
Sent: 4. november 2008 17:42
To: [email protected]
Subject: Re: [SIESTA-L] Crazy SCF with vanadium
631 atoms? whew...
Thee could be a few things that might help. One thing is to set an
electronic temperature higher than the default (about 25 meV). Even
though you say you have raised the temperature (which I assume to be
the electronic temperature), how high have you raised? Maybe you'll
have to set it to a really high value for your calculation to
converge initially, after which you'll be able to lower it
gradually, in steps of about 10 meV and re-relaxing everything again.
Another thing that helps convergence is that V atoms are in a good
initial position. My personal experience is that when the initial
positions of the atoms are way off the equilibrium ones, scf can be
problematic, especially if you have d electrons (in my case I have
Ni slabs). Supposing that V atoms would be adsorbed over the
graphene surface, you could try doing an initial relaxation with the
Harris functional - which would overcome scf convergence problems in
a first step. It will most probably not give very reliable results,
but it could help in determining an initial set of positions which
could be not too off from equilibrium. Make a test with a V atom
over a 32-atom graphene sheet to compare the outcomes of full scf
relaxation and a Harris functional calculation; if it gives decent
results, you could try doing this to your larger sheet.
Finally, I would use a larger MeshCutoff. It might help to improve
SCF convergence. Try at least 220-250 Ry.
Best of luck,
Marcos
Vous avez écrit / You have written / Lei ha scritto / Você escreveu...
Joachim Fürst
Dear all,
I am doing a relaxation with vanadium on graphene, and the scf loop
goes
crazy:
siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax
Ef(eV)
siesta: 1 -97094.38150 -99070.91635 -99070.91635 39.07032
-4.23154
timer: Routine,Calls,Time,% = IterSCF 1 882.442 95.03
elaps: Routine,Calls,Wall,% = IterSCF 1 223.031 94.74
siesta: 2 -97140.17446 -99051.35598 -99051.50902 34.72834
-4.20753
siesta: 3 -97226.76276 -98965.36619 -98965.50463 35.42815
-4.12217
siesta: 4 -94936.08243 -99371.74467 -99371.89250 35.10492
-4.22078
siesta: 5 -90678.74567 -99715.25553 -99715.47141 24.75070
-4.07217
siesta: 6 -86315.83616 -100050.89144 -100051.12562 36.83745
-3.67364
siesta: 7 -97702.64129 -98510.48955 -98510.724571051.40098
-3.89029
-And it only gets worse after more iterations.
I have tried almost all thinkable combinations of Pulay mixing,
mixing
rate etc, but without success. I have also increased temperature,
disabled spin, changed energyshift, tried different basis sets for
all
elements....No luck at all! Sometimes I can get it to converge, but
then after a CG step or two it goes nuts again.
If I leave out vanadium, the problem disappears. I have used this
vanadium PS many times in different calculations, but have never come
across this issue until now.
Have you got any ideas? I have listed my .fdf file below (the long
list of coordinates is left out).
Hope you have some good advice. Thanks!
NumberOfSpecies: 3
NumberOfAtoms: 631
LatticeConstant 1.0 Ang
%block LatticeVectors
55. 0. 0.
0. 8. 0
0. 0. 36.84
%endblock LatticeVectors
%block ChemicalSpeciesLabel
1 6 C_pbe
2 1 H
3 23 V
%endblock ChemicalSpeciesLabel
AtomicCoordinatesFormat Ang
%block AtomicCoordinatesAndAtomicSpecies
40.470458 0.000000 0.000196 1
39.049818 0.000000 0.000213 1
36.210059 -0.000001 0.000237 1
0.709653 0.000000 0.000252 1
.....
.....
%endblock AtomicCoordinatesAndAtomicSpecies
%block kgrid_Monkhorst_Pack
1 0 0 0.0
0 1 0 0.0
0 0 1 0.0
%endblock kgrid_Monkhorst_Pack
DM.UseSaveDM T
#SaveTotalPotential T
#SaveRho T
#SaveElectrostaticPotential T
#MD.UseSaveXV T
SolutionMethod Diagon
%block PAO.BasisSizes
C_pbe DZ
H DZ
%endblock PAO.BasisSizes
%block PAO.Basis
V 4 # Species label, number of l-shells
n=4 0 2 P 1 # n, l, Nzeta, Polarization,
NzetaPol
7.497 6.783
1.000 1.000
n=3 1 2 # n, l, Nzeta
3.048 1.943
1.000 1.000
n=4 1 2 # n, l, Nzeta
9.870 8.285
1.000 1.000
n=3 2 2 # n, l, Nzeta
4.901 2.758
1.000 1.000
%endblock PAO.Basis
XC.functional GGA
LongOutput .true.
XC.authors PBE
SystemLabel bulk
DM.MixSCF1 T
MaxSCFIterations 600 # Maximum number of SCF iter
DM.MixingWeight 0.01 # New DM amount for next SCF cycle
DM.Tolerance 1.d-4 # Tolerance in maximum difference
DM.UseSaveDM true # to use continuation files
DM.NumberPulay 4
SpinPolarized true
WriteMullikenPop 1
MeshCutoff 175 Ry
MD.TypeOfRun CG
MD.NumCGSteps 190
MD.UseSaveCG true
MD.UseSaveXV true
MD.MaxCGDispl 0.02 Ang
MD.MaxForceTol 0.05 eV/Ang
WriteCoorXmol true
WriteMDXmol true
WriteMDhistory true
--
Dr. Marcos Verissimo Alves
Post-Doctoral Fellow
Unité de Physico-Chimie et de Physique des Matériaux (PCPM)
Université Catholique de Louvain
1 Place Croix du Sud, B-1348
Louvain-la-Neuve
Belgique
------
Gort, Klaatu barada nikto. Klaatu barada nikto. Klaatu barada nikto.
Free translation:
Gort, Google is your friend. Google is your friend. Google is your
friend.
