Quantcast

[Arpack-ng] ARPACK question

classic Classic list List threaded Threaded
3 messages Options
Andrew Cunningham Andrew Cunningham
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

[Arpack-ng] ARPACK question

Firstly, thanks to everyone involved in maintaining ARPACK.

Maybe someone can help me with my problem. I am solving the classic 'vibration' problem in shift and invert mode (DSAUPD/DSEUPD).
In many cases, although I can store as many eigenvalues  as I could extract by writing them to disk, I do not have sufficient memory to extract them all in one operation.

Since sadly the option HOWMNY= 'S'  for DSEUPD is not implemented, I break the problem up into smaller chunks by moving the shift value around and so extract a smaller numbers of eigenvectors - that is, I actually solve the whole problem multiple times until I have all the eigenvectors  I need..

This works great...except in an extreme case where I ask for one or two eigenvalues. Then ARPACK  exceeds the number of allowed iterations. This seems odd, as if I take a case where I can extract 32 eigenvalues/vectors, then as a test, only ask for 2, the iterations are exceeded. Would anyone have an idea why this is happening?

BTW, Is anyone wokring on implementing the option HOWMNY= 'S'?

Thanks
Andrew

_______________________________________________
Arpack-ng mailing list
[hidden email]
http://lists.scilab.org/mailman/listinfo/arpack-ng
Sylvestre Ledru-4 Sylvestre Ledru-4
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

[Arpack-ng] Implementation of HOWMNY= 'S' for DSEUPD [was: Re: ARPACK question]

On 16/11/2012 21:24, Andrew Cunningham wrote:

> Firstly, thanks to everyone involved in maintaining ARPACK.
>
> Maybe someone can help me with my problem. I am solving the classic
> 'vibration' problem in shift and invert mode (DSAUPD/DSEUPD).
> In many cases, although I can store as many eigenvalues  as I could
> extract by writing them to disk, I do not have sufficient memory to
> extract them all in one operation.
>
> Since sadly the option HOWMNY= 'S'  for DSEUPD is not implemented, I
> break the problem up into smaller chunks by moving the shift value
> around and so extract a smaller numbers of eigenvectors - that is, I
> actually solve the whole problem multiple times until I have all the
> eigenvectors  I need..
>
> This works great...except in an extreme case where I ask for one or two
> eigenvalues. Then ARPACK  exceeds the number of allowed iterations. This
> seems odd, as if I take a case where I can extract 32
> eigenvalues/vectors, then as a test, only ask for 2, the iterations are
> exceeded. Would anyone have an idea why this is happening?
>
> BTW, Is anyone wokring on implementing the option HOWMNY= 'S'?
I am certainly not but I will be happy to apply any patch and make a
release with this change ;)

Sylvestre
_______________________________________________
Arpack-ng mailing list
[hidden email]
http://lists.scilab.org/mailman/listinfo/arpack-ng
Andrew Cunningham Andrew Cunningham
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Implementation of HOWMNY= 'S' for DSEUPD [was: Re: ARPACK question]

Hi,
 Modifying ARPACK at that level is beyond me , I am afraid.
 I did find a workaround to my issue. I adjust the number of eigenvectors I extract such that I never ask for <8

Andrew

On Sat, Nov 17, 2012 at 7:18 AM, Sylvestre Ledru <[hidden email]> wrote:
On 16/11/2012 21:24, Andrew Cunningham wrote:
> Firstly, thanks to everyone involved in maintaining ARPACK.
>
> Maybe someone can help me with my problem. I am solving the classic
> 'vibration' problem in shift and invert mode (DSAUPD/DSEUPD).
> In many cases, although I can store as many eigenvalues  as I could
> extract by writing them to disk, I do not have sufficient memory to
> extract them all in one operation.
>
> Since sadly the option HOWMNY= 'S'  for DSEUPD is not implemented, I
> break the problem up into smaller chunks by moving the shift value
> around and so extract a smaller numbers of eigenvectors - that is, I
> actually solve the whole problem multiple times until I have all the
> eigenvectors  I need..
>
> This works great...except in an extreme case where I ask for one or two
> eigenvalues. Then ARPACK  exceeds the number of allowed iterations. This
> seems odd, as if I take a case where I can extract 32
> eigenvalues/vectors, then as a test, only ask for 2, the iterations are
> exceeded. Would anyone have an idea why this is happening?
>
> BTW, Is anyone wokring on implementing the option HOWMNY= 'S'?
I am certainly not but I will be happy to apply any patch and make a
release with this change ;)

Sylvestre


_______________________________________________
Arpack-ng mailing list
[hidden email]
http://lists.scilab.org/mailman/listinfo/arpack-ng
Loading...