[Pw_forum] For B pseudopotential

[Stefano de Gironcoli]

A few comments

a) About machine dependency (excluding presence of bugs in the code):
1) eigenvectors of non-degenerate eigenvalues are defined up to a phase 
factor
2) eigenvectors of degenerate eigenvalues are defined up to a unitary 
rotation in the degenerate subspace
3) different machines/diagonalization routines may choose a different 
phase factors or unitary transformation
among the infinit number of  legittimate choises.

b) About bond charge models:
1) Bond charge models have been fitted to experimental frequencies, not 
to eigenvectors. These happen to be poorly described by the model if 
compared with ab-initio results. This is certainly true for GaAs but I 
bet it is so also for Ge.
2) Paolo Giannozzi worked several years ago about re-fitting bond charge 
parameters to a ab-initio eigenvalues AND eigenvectors, maybe he still 
has some data about that.

Stefano de Gironcoli

sprokuda at dnk.ru wrote:

>Dear PWscf users!
>I wanna ask you a question, which has no direct concern to PWscf.
>Generally my problem in to computer the eigenvector for bulk
>Germanium. I'mv short in time with this problem, and causeof that I've
>tried to use the code from
>O.H. Nielsen, W. Weber,     Computer Physics Communications 18 (1979) 101-108.
>
>It is based on bond-charged model and gives good results for frequencies, which are the result of the solution eigenvalue
>problem for dynamical matrix (6*6 size for Ge).
>As for eigenvectors the results seems strange and wrong. 
>First - the results of eigenvectors is strongly compilier dependent -
>that is my firsr question - how it could be and what could be reason
>for that?
>Eigenvector here is a complex vector of 6*1 size. First three digits describes eigendisplasement of first atom in unit
>cell, second three digits describes eigendisplasement of second atom. For each three e1 and e2 the condition must be
>satisfied: e1,2 = c1,2*exp(-i*phi1,2), where e1,2 is the eigendisplacement, c1,2 is a real vectors of dimention 3*1, i -
>imaginary unit.This condition is not satisfied . More in detail, in some cases this condition is satisfied,  but there only
>in some points along [q q 0] direction.
>The second question what could be the reason for such a strange behavior of comutation of eigenvectors (while the results for
>frequencies is rather good)?.
>The numerical routine for the solution of  eigenvector problem is
>well-tested, namely I use for that IMSL numerical library.
>
>I would be greatly thankfull for any information of suggestions.
>Best wishes, Sergey Prokudaylo.
>
>
>_______________________________________________
>Pw_forum mailing list
>Pw_forum at pwscf.org
>http://www.democritos.it/mailman/listinfo/pw_forum
>  
>