Giacomo,

I see you are using a mesh cutoff of 100 Ry for your calculation. This could
be too small by itself for O, which is a hard pseudo in comparison with
other elements. I believe converged calculations for  compound with O should
use at least 180 Ry (an equivalent of 45 Ry in plane waves), which is not
the most converged, either. 200 Ry is roughly equivalent to a PW 25 Ry, and
this would be good only for ultrasoft pseudos, not norm-conserving ones.

But then there is the problem of the eggbox effect. Probably with 100 Ry the
grid is so coarse you are having spurious numerical problems close to the
energy minimum. You can go to a higher cutoff to minimize this problem, and
this could even prove advantageous because your SCF will converge in less
steps, and your calculation will require less CG moves to get to the
minimum. Another "trick" is to make MaxCGDisplacement smaller than the
default. This helps locate the energy minimum more easily if the minimum is
shallow.

I don't know about the reasons why it is difficult to reproduce the
energetics of TiO2 phases, but if this is so, you might also want to go to
higher k-samplings. Are your results for the 4x4x1 grid consistent with what
you see in literature?

Cheers,

Marcos

On Thu, Jan 20, 2011 at 2:23 AM, Giacomo Giorgi <[email protected]> wrote:

> Dear All,
>        I am nowadays performing some benchmark calculations about bulk TiO2
> (rutile and anatase). The idea is to do some calculations on different
> surface reconstructions.
> I think that (I am almost sure) a pseudo for Ti with p orbitals in the
> valence is almost mandatory for getting reliable results.
>
> The problem I am facing is the following. I found a pseudo LDA for Ti
> in the website of Siesta together with the optimized basis set (Ti semicore
> states (3s and 3p) in valence)
>
> For O I use the pseudo I found in the website (Oxygen tm2, distrib with
> SIESTA, and a DZP basis set)
>
> When I test the structure of rutile (MD.Variable.Cell false), it is very
> difficult for me to get the convergence (more than 1000 CG steps). It seems
> impossible indeed to reduce the forces below 0.03 eV/Ang. This seems to me
> quite strange, also because I am restarting from a 4x4x1 Gamma centered
> previously converged structure.
> It is impossible that increasing the sampling from 4x4x1 to 4x4x2 the
> number of CG changes from 11 to more than 1000.....
> I thus think that if convergence is not achieved for this small system
> (rutile, only 2 Ti atoms) even more difficult (if not impossible) will be to
> perform calculations on slabs.
>
>
> Thus I wonder if someone can give me some suggestions in order to fix this
> problem. Is it a pseudo problem? Is it sufficient a 4e- pseudo in the
> valence for optimally describing TiO2 properties (I am quite doubtful about
> it)? I already know form literature that it is very difficult to reproduce
> the experimental energetic ordering of anatase and rutile.
>
> In case, has anyone a pseudo tested for Ti (PBE even better) used for TiO2
> based systems to share with me?
>
> SystemName          TiO2_rutilo
> SystemLabel         TiO2_rutilo
> NumberOfAtoms         6
> NumberOfSpecies       2
> %block ChemicalSpeciesLabel
>  1  22  Ti
>  2   8  O
> %endblock ChemicalSpeciesLabel
> %block PAO.Basis                 # Define Basis set
> Ti    5      1.91
>  n=3    0    1   E     93.95      5.20
>   5.69946662616249
>   1.00000000000000
>  n=3    1    1   E     95.47      5.20
>   5.69941339465994
>   1.00000000000000
>  n=4    0    2   E     96.47      5.60
>   6.09996398975307        5.09944363262274
>   1.00000000000000        1.00000000000000
>  n=3    2    2   E     46.05      4.95
>   5.94327035784617        4.70009988294302
>   1.00000000000000        1.00000000000000
>  n=4    1    1   E      0.50      1.77
>   3.05365979938936
>   1.00000000000000
> %endblock PAO.Basis
>
> %block PAO.BasisSizes
>        O      DZP
> %endblock PAO.BasisSizes
>
> XC.functional LDA
> XC.authors    CA
> LatticeConstant          4.59220 Ang
> %block LatticeVectors
>        1.0000  0.000   0.000
>        0.0000  1.000   0.000
>        0.0000  0.000   .64435346892556944384
> %endblock LatticeVectors
>
> %block kgrid_Monkhorst_Pack
>   4  0  0  0.5
>   0  4  0  0.5
>   0  0  2  0.5
> %endblock Kgrid_Monkhorst_Pack
>
> MeshCutoff         100.0 Ry
>
> MaxSCFIterations     1000
> DM.MixingWeight      0.05
> DM.NumberPulay       3
> DM.Tolerance         1.d-4
>
> SolutionMethod       diagon
>
> SpinPolarized   true
> LongOutput true
>
> MD.TypeOfRun         cg
> MD.NumCGsteps        1000
> MD.VariableCell     false #true
> MD.MaxForceTol      0.01 eV/Ang  #0.005 eV/Ang
> WriteForces             true
> WriteCoorCerius         true
> WriteCoorXmol           false
> WriteDenchar            true
> WriteMullikenPop        1
>
>
> UseSaveData      true
>
> Diag.ParallelOverK  true
>
> AtomicCoordinatesFormat Fractional
> %block AtomicCoordinatesAndAtomicSpecies
>    0.00000   0.00000   0.00000         1
>    0.50000   0.50000   0.50000         1
>    0.30510   0.30510   0.00000         2
>    0.19490   0.80510   0.50000         2
>    0.69490   0.69490   0.00000         2
>    0.80510   0.19490   0.50000         2
> %endblock AtomicCoordinatesAndAtomicSpecies
>
>
>
> .....
> siesta:                 ==============================
>                            Begin CG move =   1000
>                        ==============================
>
> outcoor: Atomic coordinates (fractional):
>   -0.00002412    0.01663451   -0.00906965   1  Ti         1
>    0.49999230    0.51667694    0.49093035   1  Ti         2
>    0.29977119    0.33204697   -0.00156798   2  O          3
>    0.20023279    0.83205355    0.49848870   2  O          4
>    0.70023510    0.70128165   -0.00155817   2  O          5
>    0.79979393    0.20130016    0.49847493   2  O          6
>
> superc: Internal auxiliary supercell:     4 x     4 x     7  =     112
> superc: Number of atoms, orbitals, and projectors:    672 10080 11648
>
> outcell: Unit cell vectors (Ang):
>        4.592200    0.000000    0.000000
>        0.000000    4.592200    0.000000
>        0.000000    0.000000    2.959000
>
> outcell: Cell vector modules (Ang)   :    4.592200    4.592200    2.959000
> outcell: Cell angles (23,13,12) (deg):     90.0000     90.0000     90.0000
> outcell: Cell volume (Ang**3)        :     62.4003
>
> siesta: iscf   Eharris(eV)      E_KS(eV)   FreeEng(eV)   dDmax  Ef(eV)
> siesta:    1    -4921.3185    -4921.3185    -4921.3185  0.0000 -4.1784
>
> siesta: E_KS(eV) =            -4921.3185
>
> siesta: Atomic forces (eV/Ang):
>     1    0.001276    0.003726   -0.037969
>     2   -0.000729   -0.003555   -0.039057
>     3    0.003563    0.007823    0.004704
>     4   -0.006634    0.009005   -0.003266
>     5   -0.003276   -0.006490    0.003888
>     6    0.000690   -0.007992    0.002003
> ----------------------------------------
>   Tot   -0.005110    0.002516   -0.069698
> ----------------------------------------
>   Max    0.039057
>   Res    0.013671    sqrt( Sum f_i^2 / 3N )
> ----------------------------------------
>   Max    0.039057    constrained
>
>
> Thanks in advance!
> Best
>
> Giacomo
>
>
>
>
>
>
> * * * * * * * * * * * * * * * * * * * * * * * * *
> *   Giacomo Giorgi                              *
> *   Ph.D.                                       *
> *   Dipartimento di Chimica e Istituto di       *
> *   Scienze e Tecnologie Molecolari ISTM-CNR    *
> *   Universita' degli Studi di Perugia          *
> *   Via Elce di Sotto 8  06123 Perugia Italia   *
> *   Tel: +39 075-5855618  Fax: +39 075-5855606  *
> *   E-mail: giac_at_thch.unipg.it               *
> * * * * * * * * * * * * * * * * * * * * * * * * *
>
>
>
>
>
>
> "Oltre le illusioni di Timbuctu'
> e le gambe lunghe di Babalu' c'era questa strada...
> Questa strada zitta che vola via come una farfalla,
> una nostalgia, nostalgia al gusto di curacao...
> Forse un giorno meglio mi spieghero'..."
>
> (Paolo Conte, "Hemingway")
>
>

Responder a