[Pw_forum] Fwd: Re: Re: GIPAW memory problem?

[Davide Ceresoli]

Dear Aleksander,
     in this moment I have only 2 CPUs for debugging, and it's
going really slow.

I suspect that the problem is due to the recent changes to
GIPAW/src/compute_u_kq.f90, in how u_{k+q} is initialized from
the wavefunction at k. I think that the orthogonalization
is lost here.

Maybe the others developers can help me, by telling if this
is the proper way to diagonalize the hamiltonian at fixed
KS potential.

Best wishes,
     Davide




On 06/03/2014 10:11 AM, Paolo Giannozzi wrote:
> On Mon, 2014-06-02 at 22:39 +0200, Aleksander Jaworski wrote:
>
>> I would like to ask you if there are any improvements appearing
>> on the horizon regarding my gipaw problems :)
>
> if the overlap matrix is not positive definite (and if the code
> says it isn't, it isn't), the solution of the secular equation
> (H-Se)\psi=0 is not possible, unless you find a way to keep
> eigenvectors of S with nonpositive eigenvalues out of business.
>
> P.
>
>> best,
>> Aleksander