PARPACK issues

34 views
Skip to first unread message

Ehsan Mihankhah

unread,
Sep 19, 2016, 3:04:35 AM9/19/16
to deal.II User Group
Hi,

I have been working with Dealii 8.3 for a while and I have been using the modified version of step-36 to solve eigenvalue problem with ARPACK. I noticed that there is a possibility of working with PARPACK in Dealii 8.4 and so I upgraded to this version.
I am trying to get the codes in
https://github.com/dealii/dealii/tree/master/tests/arpack
running with MPI and PETSc variables.
I managed to compile the sample codes but the problem is that when I run the code I receive errors from ARPACK referring to "info=-9" in Pdnaupd. and following "info=-9" I see the error is about initial vector being zero and I cannot solve this problem by tuning any parameter it seems.
Can anybody suggest a solution please?

Thank you very much,
Best regards,
Ehsan,

P.S: One important issue here is that the parpack_solver.h that comes with the package available on dealii website is not the one on the github and therefore it does not recognize some of the functions. So, I copied the version which is available on the github in order to be able to get the codes compiled.

Denis Davydov

unread,
Sep 20, 2016, 2:16:24 AM9/20/16
to deal.II User Group
Hi Ehsan,


On Monday, September 19, 2016 at 9:04:35 AM UTC+2, Ehsan Mihankhah wrote:
Hi,

I have been working with Dealii 8.3 for a while and I have been using the modified version of step-36 to solve eigenvalue problem with ARPACK. I noticed that there is a possibility of working with PARPACK in Dealii 8.4 and so I upgraded to this version.
I am trying to get the codes in
https://github.com/dealii/dealii/tree/master/tests/arpack
running with MPI and PETSc variables.

So you are combining a installation of 8.4.x with unit tests from the current master branch.
This is generally a bad idea. I would suggest you configure the current development version of deal.II from GtiHub, compile and then
try to run those tests by:

make setup_tests_arpack
ctest -R "arpack"

 
I managed to compile the sample codes but the problem is that when I run the code I receive errors from ARPACK referring to "info=-9" in Pdnaupd. and following "info=-9" I see the error is about initial vector being zero and I cannot solve this problem by tuning any parameter it seems.
Can anybody suggest a solution please?

Thank you very much,
Best regards,
Ehsan,

P.S: One important issue here is that the parpack_solver.h that comes with the package available on dealii website is not the one on the github and therefore it does not recognize some of the functions. So, I

that's not an issue, but the expected behaviour. Deal.II 8.4.x is the 8.4.x version of the code, whereas "master" branch is the current development version. The two are not the same by definition.
 
copied the version which is available on the github in order to be able to get the codes compiled.

 
Regards,
Denis

Denis Davydov

unread,
Sep 20, 2016, 2:18:55 AM9/20/16
to deal.II User Group
p.s. Alternatively, you can check those tests in 8.4. https://github.com/dealii/dealii/tree/dealii-8.4 branch

Denis Davydov

unread,
Sep 20, 2016, 3:05:09 AM9/20/16
to deal.II User Group


On Monday, September 19, 2016 at 9:04:35 AM UTC+2, Ehsan Mihankhah wrote:
I managed to compile the sample codes but the problem is that when I run the code I receive errors from ARPACK referring to "info=-9" in Pdnaupd. and following "info=-9" I see the error is about initial vector being zero and I cannot solve this problem by tuning any parameter it seems.

if you set initial vector by set_initial_vector(), then make sure it's nonzero (i.e. random vector or a solution from previous calculations).
Otherwise a random vector will be used internally by (p)Arpack, see

Regards,
Denis.

 
Reply all
Reply to author
Forward
0 new messages