Maxim,

Your restart procedure is flawed because the structure of the DM is
not the same in the Gamma point calculation and in the k-point
calculation. When using k-points, Siesta generates an auxiliary
supercell, and the orbital indexes are referred to it, instead of to
the unit cell.
You can see this in your output file for the k-point calculation:

initatomlists: Number of atoms, orbitals, and projectors:     28   364   448
....
superc: Internal auxiliary supercell:     4 x     2 x     4  =      32
superc: Number of atoms, orbitals, and projectors:    896  11648  14336


A bit further on in your output file for the k-point calculation, the
diagnostic of a different structure for the DM can be seen clearly in
the lines:

New_DM. Step:     1
Initializing Density Matrix...

iodm: Reading Density Matrix from files
 Read DM has different structure. Fixing...

The "fixing" step is really meant for small-scale structural changes,
such as those ocurring between MD steps. In this case, the "fixing" is
hopeless and would result in an starting DM that is way off from your
convergence basin.

For this kind of "mixed-mode" restart, I would recommend using the
charge-density reading options documented in the manual.

  Regards,

   Alberto

2011/10/20 Максим В. Песков <[email protected]>:
> Hi Siesta folks,
>
>  I wonder if somebody can give me a hint about how to restart from
>  density matrix generated at a lower kpoint set. I took a simple
>  example of silicon carbide-nitride to demonstrate the problem.
>
>  First of all, I run geometry optimization at gamma point using the
>  input file Si2CN4.fdf (attachment). The output for this run you can
>  see in output1.log file.
>  Now I take files ceramic.CG, ceramic.DM and ceramic.XV from gamma
>  point run and setup another simulation where k-sampling is 4x2x4.
>  The following changes have been made to Si2CN4.fdf:
>
>  MD.NumCGsteps             0
>  MD.UseSaveXV              T
>  MD.UseSaveCG              T
>  DM.UseSaveDM              T
>  DM.MixSCF1                T
>  MaxSCFIterations          0
>
>  %block kgrid_Monkhorst_Pack
>  4   0   0   0.0
>  0   2   0   0.0
>  0   0   4   0.0
>  %endblock kgrid_Monkhorst_Pack
>
>  Output of the run with a higher kpoints has large atomic forces on
>  some atoms, what indicates that  density matrix  has
>  non-consistent data (full output in the file output1_4x2x4).
>
>  Am I doing something wrong? Where is the catch?
>
>  siesta: Atomic forces (eV/Ang):
>  siesta:      1    2.317829   23.272352   -0.184196
>  siesta:      2   -4.781680   41.922932    0.808236
>  siesta:      3    0.094806   -0.139280    1.621746
>  siesta:      4    8.002834   69.587296    2.341780
>  siesta:      5   -1.288002   -8.407007   -1.907456
>  siesta:      6   -7.220729    4.432225   14.978416
>  siesta:      7    4.520995   -1.714879   11.935076
>  siesta:      8    4.747628   -4.123018    7.092739
>  siesta:      9   -4.009865    9.527266    7.324178
>  siesta:     10   -3.822497  -22.299940   -2.265516
>  siesta:     11    1.828340  -29.266963   -4.327223
>  siesta:     12    5.854673  -28.756784   -4.665025
>  siesta:     13   -3.615627   23.764781   -1.472908
>  siesta:     14    3.512788  -24.591626   -1.517415
>  siesta:     15    1.060980   -5.722265    0.855264
>  siesta:     16    0.741338   -0.062121    9.799283
>  siesta:     17   -1.379791   -0.654582   14.161918
>  siesta:     18   -4.012675   -0.552881   12.470952
>  siesta:     19    3.747949   -0.198465   11.466282
>  siesta:     20   -7.703283  -59.798647   -3.272396
>  siesta:     21    7.239848   -0.696221  -10.859396
>  siesta:     22   -3.416753    9.754062   -7.922024
>  siesta:     23   -4.963762   -0.880779   -7.061494
>  siesta:     24    3.234729   12.479415   -5.571744
>  siesta:     25    2.512698   -1.670932   -6.844461
>  siesta:     26   -2.612930   -0.070914   -8.507862
>  siesta:     27   -0.273650   -9.408214  -16.537933
>  siesta:     28   -0.315865    4.294655  -11.942198
>
>  --
>  Thanks,
>  Max

Responder a