[Q-e-developers] threshold for diagonalization

[stefano de gironocoli]

I think the idea was that the threshold is set such that the band energy 
(the sum of the eigenvalues) is converged well within  tr2...
where "well within" means apparently a factor of 10

this is the recipe used in the scf calculation as well because you don't 
want the scf error to be spoiled by the error in the eigenvalue sum.

stefano

On 06/21/2013 11:49 AM, Paolo Giannozzi wrote:
> I just noticed that there is an inconsistency
> between what the documentation says about the
> threshold for iterative diagonalization in
> non-self-consistent calculations:
>    For non-scf calculations: default is conv_thr/N elec .
> and what is actually used (1/10 of the above)
>    IF ( ethr == 0.D0 ) ethr = 0.1D0 * MIN( 1.D-2, tr2 / nelec )
> Any comments on this choice? it seems to me a little
> bit overzealous
>
> P.