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: SIESTA-L@listserv.uam.es 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.