[Pw_forum] optimizing pw input parameters

[Paolo Giannozzi]

On Sunday 23 December 2007 09:23, meisam aghtar wrote:

> I have to mention that there is overlap in davidson diagonalization.

of course there is. The davidson algorithm generates a subspace of 
trial and correction vectors and performs a generalized diagonalization 
in the subspace: h v_i = e_i s v_i, where s is the "overlap matrix" 
between trial+correction vectors. In the past there used to be another
version of the algorithm that used a "normal" diagonalization instead: 
h v_i = e_i v_i . This is the reason for the message.

Paolo
-- 
Paolo Giannozzi, Democritos and Udine University