Friday 07 Jan 2011
I've done the phonon sample calculations for Si at Gamma and X with both Normconserving and Ultrasoft pseudopotential. There appears to be a fair amount of variation. This variation is more likely to stem from the use of a ridiculously low cut-off, different exchange-correlation functionals and different pseudopotentials. So wouldn't read too much in to this. 

The point is the PH code runs for Norm conserving and Ultra soft (as advertised).   

NC- Si.pz-vbc.UPF
Exchange-correlation      =  SLA  PZ   NOGX NOGC (1100)
     Diagonalizing the dynamical matrix
     q = (    1.000000000   0.000000000   0.000000000 )
 **************************************************************************
     omega( 1) =       4.209763 [THz] =     140.423499 [cm-1]
     omega( 2) =       4.209763 [THz] =     140.423499 [cm-1]
     omega( 3) =      12.237902 [THz] =     408.215189 [cm-1]
     omega( 4) =      12.237902 [THz] =     408.215189 [cm-1]
     omega( 5) =      13.745644 [THz] =     458.508382 [cm-1]
     omega( 6) =      13.745644 [THz] =     458.508382 [cm-1]
 **************************************************************************

For Ultrasoft Pseudopotentials:
Ultrasoft - Si.pbe-van_gipaw.upf
SLA  PW   PBX  PBC (1434)
     q = (    1.000000000   0.000000000   0.000000000 )
 **************************************************************************
     omega( 1) =       4.143612 [THz] =     138.216954 [cm-1]
     omega( 2) =       4.143612 [THz] =     138.216954 [cm-1]
     omega( 3) =      12.393498 [THz] =     413.405357 [cm-1]
     omega( 4) =      12.393498 [THz] =     413.405357 [cm-1]
     omega( 5) =      14.010465 [THz] =     467.341916 [cm-1]
     omega( 6) =      14.010465 [THz] =     467.341916 [cm-1]
 **************************************************************************

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

For Norm conserving case
Si.pz-vbc.UPF
SLA  PZ   NOGX NOGC (1100)
     q = (    0.000000000   0.000000000   0.000000000 )
**************************************************************************
     omega( 1) =       0.094459 [THz] =       3.150827 [cm-1]
     omega( 2) =       0.094459 [THz] =       3.150827 [cm-1]
     omega( 3) =       0.094459 [THz] =       3.150827 [cm-1]
     omega( 4) =      15.294065 [THz] =     510.158489 [cm-1]
     omega( 5) =      15.294065 [THz] =     510.158489 [cm-1]
     omega( 6) =      15.294065 [THz] =     510.158489 [cm-1]
**************************************************************************

For Ultra soft case
Si.pbe-van_gipaw.upf
SLA  PW   PBX  PBC (1434)
     q = (    0.000000000   0.000000000   0.000000000 )
**************************************************************************
     omega( 1) =       0.355716 [THz] =      11.865491 [cm-1]
     omega( 2) =       0.355716 [THz] =      11.865491 [cm-1]
     omega( 3) =       0.355716 [THz] =      11.865491 [cm-1]
     omega( 4) =      15.431037 [THz] =     514.727411 [cm-1]
     omega( 5) =      15.431037 [THz] =     514.727411 [cm-1]
     omega( 6) =      15.431037 [THz] =     514.727411 [cm-1]
**************************************************************************

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Jan 10 2010 Now some GW stuff. I am running some debugs to see what is wrong with my calculation of the static screened coulomb interaction. 

From SGW a non scf calculation gives for parameters (5 ryd cutoff,5by5by5 kpoint sampling):
    Non SCF Screened Coulomb: q =  0.010  0.000  0.000  G =  0.000  0.000  0.000  w(eV) =  0.000
    0.00000       1.00000  0.00000 
    w(iw)          dvbare(nl(1))

    15.80805  0.00000
    ^^^^^^^^^^^^^^^^^^^  
    eps_{00}(0.01, 0)  

For 5 ryd cutoff 6x6x6 sampling we get
    Non SCF Screened Coulomb: q =  0.010  0.000  0.000  G =  0.000  0.000  0.000  w(eV) =  0.000
    0.00000       1.00000  0.00000      13.88798  0.00000

And a fully self-consistent calculation gives:
    inveps_{GG}(q,w) =   0.06977  0.00000
    (1/0.06977) = 14.33

for an 8x8x8 2 ryd cutoff 
        inveps_{GG}(q,w) =   0.09127  0.00000

for an 8x8x8 5 ryd cutoff
        inveps_{GG}(q,w) =   0.09108  0.00000

for a 5x5x5 grid 2 ryd:
    inveps_{GG}(q,w) =   0.06967  0.00000

5x5x5 5ryd grid
    inveps_{GG}(q,w) =   0.06977  0.00000

So it looks like 6x6x6 gives the best inveps.

for 6x6x6 2 ryd cutoff
        inveps_{GG}(q,w) =   0.07970  0.00000

6x6x6 grid 5 ryd gives:
        inveps_{GG}(q,w) =   0.07971  0.00000

The kpoint sampling seems to effect the calculated values much more than the number of gvectors. In fact it is essentially the same for 5+ rydbergs. Now what am I calculating in QuesGw?