[Pw_forum] cp.x and lemieux TRU64 Alpha

Silviu Zilberman silviu at Princeton.EDU
Tue Oct 26 17:56:30 CEST 2004


Maybe you try using steepest descent for some number of steps, and then 
move back to the 'ortho'?


Yonas Abraham wrote:

>Hi Paolo,
>It looks like the 'bug' is machine dependent. we took
>the same input parameter identical source code and run
>it in PGI compiler both 32 and 64 bit versions and
>intel on 32 bit machines and worked just fine. even
>though small perturbation in the machine accuracy
>difference can create instability, i would expect the
>difference would be bigger between 32 and 64 bits. 
>here is the snapshot of the otho diff.
> diff=    2.58117779926850       iter=            1
> diff=    6.73012657805981       iter=            2
>  ......///....
> diff=   5.958041775915444E+117  iter=           10
> diff=   1.644433412630960E+235  iter=           11
>then it crushes.
>we have tried different dt values and didn't help.
>any help would be appreciated
>--- Paolo Giannozzi <giannozz at nest.sns.it> wrote:
>>On Tuesday 26 October 2004 14:54,
>>aaron at nemo.physics.ncsu.edu wrote:
>>>We have isolated the failure to the subroutine
>>does the failure occur in one of the first steps
>>after a 
>>restart? the iterative orthonormalization is
>>unstable if psi(t+dt) and psi(t) differ too much.
>>Try to
>>reduce the time step, maybe just for the first few
>>Paolo Giannozzi             e-mail: 
>>giannozz at nest.sns.it
>>Scuola Normale Superiore    Phone:   +39/050-509876,
>>Piazza dei Cavalieri 7      I-56126 Pisa, Italy
>>Pw_forum mailing list
>>Pw_forum at pwscf.org
>Do you Yahoo!?
>Yahoo! Mail Address AutoComplete - You start. We finish.
>Pw_forum mailing list
>Pw_forum at pwscf.org

Zilberman Silviu
213 Frick Laboratory, 
Department of Chemistry, 
Princeton University
Princeton, NJ 08544, USA
phone: 609-258-1834
fax:   609-258-6746
silviu at Princeton.EDU

More information about the users mailing list