[Pw_forum] error in rdiaghg

Paolo Giannozzi giannozz at nest.sns.it
Mon Oct 6 22:55:08 CEST 2003


On Sunday 05 October 2003 21:39, Eyvaz Isaev wrote:
> Dear Sergei,
>
> You can try to change the diagonalization key, i.e.
> "cg" instead of davidson.

I don't think he can: "rdiaghg" is used in the Gamma-only
version of the code, which uses the trick to perform two 
FFTs at the same time. Since the conjugate-gradient type
diagonalization operates on one band at the time, it cannot 
work with two FFTs (on two bands) at the same time, unless 
some nontrivial change is made to the algorithm (everybody
is welcome to try).

Coming to the original question: in the manual there are some
possible reasons for the similar error in routine "cdiaghg",
which apply as well to this case. It seems to me however
that the case shown is a case of failure of the lapack
diagonalization routines used in *diaghg. Whether this is
a problem of lapack (unlikely) or a problem of a specific
machine-dependent implementation of lapack (more likely)
or a compiler problem or some subtle problem in the 
code (typically, memory overrun), it is hard to say.
Some time ago I spent several hours trying to pinpoint the
origin of a similar problem with cdiaghg on a sp3, to no avail.

So, the only advice I can offer is to restart the calculation 
from where it crashed and see what happens

Paolo

--
Paolo Giannozzi             e-mail:  giannozz at nest.sns.it
Scuola Normale Superiore    Phone:   +39/050509412
Piazza dei Cavalieri 7      Fax:     +39/050509417, 050563513
I-56126 Pisa, Italy         Office:  Lab. NEST, Via della Faggiola 19



More information about the users mailing list