Piotr, One possible reason is that your convergence parameters are not sufficiently strict. DM convergences of 10^-3 will give you very poor results. Ideally, the DM convergence should be as strict as possible, and you can easily get down to 10^-5 or even 10^-6 with systems as easy to handle as graphene.
However, perhaps the most important thing is to check that Harris and Kohn-Sham energies are equal. Check Eq. 61 of the Siesta paper (the one with the description of the code, http://es.arxiv.org/abs/cond-mat/0111138). Thus, set the DM tolerance low enough so that the Harris and KS energies are equal at the end of your SCF cycle, which will improve the numerical accuracy of the forces and stresses on the system. Your SCF cycles will be longer, but it could be that you get to your target stress tensor in less CG steps. Also, one thing that might help you is to set the off-diagonal and zz stress-tensor components to zero at the end of each step, since they should be just numerical noise due to a finite grid fineness. You can check the manual for the use of subroutine constr and the setting of stress tensor elements there. Only after you get reliable forces and stresses for a given mesh cutoff (with good scf convergence) should you worry about how it increases with a change in the mesh cutoff. Cheers, Marcos On Thu, Sep 15, 2011 at 12:31 AM, Piotr Tempczyk <piotrtempc...@wp.pl>wrote: > ** > Dear siesta users, > > I have a problem with obtaining a graphene 2x2 cell (8 atoms) with > Stress-Tensor-Voigt elements less than 0.01 kBar. > The best I have managed to get is: > Stress-tensor-Voigt (kbar): 0.02 0.01 0.00 > 0.01 0.00 0.00 > with fdf attached at the end of this email. > > I was trying to obtain it with parameters > > MD.VariableCell T > MD.MaxStressTol 0.001 GPa > > but siesta made 1000 CG steps and did not manage to achieve a goal. > I have also discovered, that when I incerase MeshCutoff from 300 to 400 Ry > the stress > tensor elements become ten times bigger. Why is that happening? > > In my future work I want to replace one or more carbon atoms in my 2x2 cell > with > other atoms, so I need to optimize that heterogenous cell and I am afraid, > that > it will be much more complicated problem that this with pure graphene, so I > will > appreciate any hints how to optimize pure graphene and that with other > atoms. > > I will be grateful for any help and hints about my problem, > > Best regards, > Peter Tempczyk > > University of Warsaw > ul. Hoza 69 > Pl-00-681 WARSZAWA, POLAND > > > > > SystemLabel graf > NumberOfAtoms 8 > NumberOfSpecies 1 > > %block ChemicalSpeciesLabel > 1 6 C > %endblock ChemicalSpeciesLabel > > PAO.BasisSize > DZP > > > PAO.BasisType split > > kgrid_cutoff 10 Ang > > MeshCutoff 300. Ry > > SolutionMethod diagon # diagonalizacion. order-N. TD-DFT > > MaxSCFIterations 1000 # Max. number of iterations in SCF. > DM.MixingWeight 0.1 # weight of mixing in the DM for > convergence > DM.Tolerance 1.d-3 # Tolerance in differences of DM > DM.NumberPulay 4 # Number of previous steps of DM for > mixing > ElectronicTemperature 300 K > MD.TypeOfRun CG #molecular dynamics conjugate gradients > MD.NumCGsteps 1000 # number of steps > MD.MaxForceTol 0.00001 eV/Ang > MD.MaxCGDispl 0.01 Bohr > DM.Energy.Tolerance 1.d-3 eV # Tolerance in energy > DM.Require.Harris.Convergence T > DM.Harris.Tolerance 1.d-3 eV > DM.Require.Energy.Convergence T # energia musi zejsc ponizej zadanego > poziomu!!! > > > WriteCoorInitial T > WriteCoorStep T # must be set true to write coords at each > relax step to .ANI file > WriteForces T > WriteKpoints F > WriteEigenvalues F # use with eig2dos to plot density of > states > WriteKbands F > WriteBands F > WriteCoorXmol T > WriteCoorCerius T # writes final coordinats for Cerius > WriteMDCoorXmol T > WriteDM T # true is default - allows for a restart > using DM from previous run > WriteMDhistory T > WriteCoorXmol T > # Options for saving/reading information > DM.UseSaveDM T # Use DM Continuation files (use for > restarting jobs) > MD.UseSaveXV T # Use stored positions and velocities (use > for restart jobs) > UseSaveData T # useful if a restart is needed > MD.UseSaveCG T # Use stored positions and velocities > SaveRho F # Write valence pseudocharge at the mesh > SaveDeltaRho F # Write RHOscf-RHOatm at the mesh > SaveElectrostaticPotential F # Write the total elect. pot. at the mesh > SaveTotalPotential F # Write the total pot. at the mesh > WriteSiestaDim F # Write minimum dim to siesta.h and stop > WriteDenchar F # Write information for DENCHAR > NetCharge 0.00 > #SlabDipoleCorrection T > #SimulateDoping T > > SpinPolarized F # spin polarization > xc.functional GGA # flavour of functional (LDA/GGA) > xc.authors PBE # Parametrization fos xc > > WriteMullikenPop 0 > WriteEigenvalues .true. > WriteWaveFunctions .true. > > #inny sposob podawania parametrow > LatticeConstant 2.481807 Ang > %block LatticeParameters > 2.000000 2.000000 60.0 90. 90. 60. > %endblock LatticeParameters > > AtomicCoordinatesFormat NotScaledCartesianAng > > %block AtomicCoordinatesAndAtomicSpecies > 1.24192412 0.71702548 0.00000 1 > 2.48283025 1.43346309 0.00000 1 > 2.48282797 2.86633383 0.00000 1 > 3.72373434 3.58277111 0.00000 1 > 3.72373167 0.71702567 0.00000 1 > 4.96463747 1.43346348 0.00000 1 > 4.96463527 2.86633370 0.00000 1 > 6.20554162 3.58277150 0.00000 1 > %endblock AtomicCoordinatesAndAtomicSpecies > > AtomCoorFormatOut Ang > >