[Pw_forum] EXX routine

[Helen]

Thank you for your reply,

> I'll indeed let Stefano chip in, or any other - one question that comes
> to mind is this: would the results change if you had a much tighter
> tolerance for convergence ?
>
> The code has "global" tolerance thresholds, to which the quasi-empty
> k-points do not contribute. I think the Davidson algorithm should
> diagonalize equally well the quasi-empty, but if it struggles, the code
> will not recognize their struggle.

I had tried increasing convergence tolerances but it didn't help

> A few more points: I am not sure with which precision the
> code reads the weights (maybe your small weights are read as
> zero) - and what it does if the sum of the weights
> is not normalized to 1.0 or 2.0 (as in your case).

I checked this within the EXX code, and it is OK

> Last sanity check: what happens if you also give k=0,0,0
> a second time, but with the almost-zero weight ?
>
This is a very good idea, I have just checked it, with strange results, when 
I do the normal 8 k-points and add at the end k=0,0,0 weight=0.0000001 then 
I get the correct results. But when I add at the end both k=0,0,0 and 
k=0.5,0.5,0.5 and k=0,0,1 (the symmetry points needed for nq=2,2,1) all with 
negligible weights I get wrong results for the orbital energies of k=0,0,0. 
I will look more closely why this may be happening,
Thanks for your help,
Helen

Fritz-Haber Center, HebrewUniversity