[petsc-users] SLEPC solvers not converging in latest version

[Pandey, Siddhant]

I've attached the B matrix (binary) in a zip file. To give more context to the problem, the previous installation of petsc/slecp doesn't give this error. So the matrix being singular shouldn't be the issue since my matrix assembly routines are independent of the petsc/slepc installation I use. Also, diagonalization happens for smaller matrices.


Unfortunately, since EPSsolve never got completed, -eps_viewer didn't return anything. Is there any other way I can show the settings used?


Sidd

________________________________
From: Jose E. Roman <jroman at dsic.upv.es>
Sent: Wednesday, July 12, 2017 2:02:55 PM
To: Hong Zhang
Cc: petsc-users at mcs.anl.gov; Pandey, Siddhant
Subject: Re: [petsc-users] SLEPC solvers not converging in latest version

Let me correct my last message: we set this option in the development version (master), so yes, in 3.7 it may be good to try -ksp_error_if_not_converged
Sorry for the noise.
Jose


> El 12 jul 2017, a las 19:58, Jose E. Roman <jroman at dsic.upv.es> escribió:
>
> Hong: in the latests versions of SLEPc we turn on this option by default (except in some solvers such as JD), so I don't think the problem is in the KSP solve.
> Jose
>
>
>> El 12 jul 2017, a las 19:53, Hong <hzhang at mcs.anl.gov> escribió:
>>
>> Add option '-ksp_error_if_not_converged'.  I suspect the matrix has zero pivot.
>> Hong
>>
>> On Wed, Jul 12, 2017 at 12:45 PM, Jose E. Roman <jroman at dsic.upv.es> wrote:
>>
>>> El 12 jul 2017, a las 18:56, Pandey, Siddhant <spandey2 at wpi.edu> escribió:
>>>
>>> We recently updated our installation of PETSC and SLEPC from an older version, and eigenvalue problems converging in the previous installation is not converging anymore (diagonalization never comes to an end). On using -info, the following message keeps repeating during diagonalization:
>>>
>>> [0] PCSetUp(): Leaving PC with identical preconditioner since operator is unchanged
>>> [0] STMatSolve(): Linear solve iterations=1
>>> [0] PCSetUp(): Leaving PC with identical preconditioner since operator is unchanged
>>> [0] STMatSolve(): Linear solve iterations=1
>>> [0] BV_SafeSqrt(): Zero norm, either the vector is zero or a semi-inner product is being used
>>> [0] PCSetUp(): Leaving PC with identical preconditioner since operator is unchanged
>>> [0] STMatSolve(): Linear solve iterations=1
>>> [0] PCSetUp(): Leaving PC with identical preconditioner since operator is unchanged
>>> [0] STMatSolve(): Linear solve iterations=1
>>>
>>> As can be seen, iteration number remains 1. The issue gets resolved with a smaller sized matrix. Currently I am using a 10648 x 10648 matrix, and PETSC is configured for complex values.
>>>
>>> What changes do I have to make to make my codes work in the new installation? This seems to be a version specific issue.
>>>
>>>
>>> Thanks,
>>> Sidd
>>> WPI
>>
>> I don't understand why are you getting the BV message. Maybe your B-matrix is singular, but it is not sufficient to explain this. If possible, send me your matrices and I will try to reproduce the problem (I would also need to know all the settings, which can be displayed with -eps_view or -eps_view_pre).
>>
>> Jose
>>
>>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20170712/dd9702fe/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: B_mat_binary.zip
Type: application/zip
Size: 853677 bytes
Desc: B_mat_binary.zip
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20170712/dd9702fe/attachment-0001.zip>