# [Trilinos-Users] Beginner's problems with PyTrilinos and Anasazi

Marko Budisic mbudisic at engineering.ucsb.edu
Tue Dec 16 12:10:46 MST 2008

```Chris,

I am sorry, I made a typo in my reply. I have rescaled matrix rows with
_row_ sums (not matrix sums). And your computations are right for W
prior to rescaling. After rescaling, the eigenvalues are as posted
before.

Best,
Marko

On Tue, 2008-12-16 at 14:06 -0500, Marko Budisic wrote:
> Hi Chris,
>
> those eigenvalues are right for the matrix W. However, if you look back
> into the code, you'll see that immediately after populating the entries
> of the matrix, I rescale the matrix rows with matrix sums to create a
> row-stochastic matrix. After that eigenvalues are those that I've
> included in my reply to Heidi.
>
> Thanks,
> Marko Budisic
>
>
> On Tue, 2008-12-16 at 11:56 -0700, Baker, Christopher Grover wrote:
> > Marko,
> >
> > If I am correct in reading the code below, the matrix A is:
> > W =
> >    (1,1)        1
> >    (1,2)        2
> >    (1,3)        3
> >    (1,5)        2
> >    (2,2)        1
> >    (2,3)        2
> >    (3,3)        1
> >    (3,4)        2
> >    (4,1)        3
> >    (4,4)        1
> >    (5,4)        2
> >    (5,5)        1
> >
> > or, alternatively:
> >
> >      1     2     3           2
> >             1     2
> > W =               1     2
> >       3                 1
> >                         2     1
> >
> > According to MATLAB eig() and eigs(), the eigenvalues of this matrix
> > are:
> >   4.3379
> >   -0.2754 + 2.7399i
> >   -0.2754 - 2.7399i
> >    0.2128
> >    1.0000
> >
> > These are not the same eigenvalues that you list below. Do I have the
> > matrix wrong?
> >
> > Chris
> >
> > On 12/16/08 11:28 AM, "Marko Budisic" <mbudisic at engineering.ucsb.edu>
> > wrote:
> >
> >         Hi Heidi,
> >
> >
> >         Indeed, I have intended to create a 5x5 matrix and the code
> >         a faulty code, probably left over from when I was trying to
> >         play with
> >         parameters to see if I understood them properly. The line
> >
> >         numglobals = nx*nx
> >
> >         should be changed to
> >
> >         numglobals = nx
> >
> >         to get a matrix with 5 rows. The matrix population code is
> >         complete.
> >         However, this does not resolve the original problem. For
> >         information, here are eigenvalues of the matrix that I would
> >         like to
> >         compute (as computed by numpy and numpy.linalg):
> >
> >         1.00000000+0.j
> >         -0.01363536+0.59362319j
> >         -0.01363536-0.59362319j
> >         0.06893738+0.j
> >         0.33333333+0.j
> >
> >         Thank you for your help.
> >
> >         Best,
> >         Marko
> >
> >
> >         On Tue, 2008-12-16 at 11:09 -0700, Thornquist, Heidi K wrote:
> >         > Hi Marko,
> >         >
> >         > Thanks for sending a copy of your code.  Upon inspection of
> >         the
> >         > attached code snippet, it looks like you
> >         > are creating a matrix with 25 rows and columns and then only
> >         filling
> >         > the primary 5x5 block.  This gives
> >         > the matrix a nullspace of dimension 20.  Is this the matrix
> >         you meant
> >         > to generate, or did you omit some
> >         > of the matrix population for brevity?  If so, could you
> >         include all
> >         > the matrix population code.
> >         >
> >         > Thanks,
> >         > Heidi
> >         >
> >         > ---
> >         > Heidi K. Thornquist
> >         > Senior Member of Technical Staff
> >         > Electrical & Microsystem Modeling
> >         > Sandia National Laboratories
> >         > P.O. Box 5800, MS 0316
> >         > Albuquerque, NM  87185-0316
> >         > Phone: (505)284-8426
> >         > Fax:   (505)284-5451
> >         >
> >         >
> >         >
> >         >
> >         > On 12/15/08 9:47 PM, "Marko Budisic"
> >         <mbudisic at engineering.ucsb.edu>
> >         > wrote:
> >         >
> >         >         Dear Trilinos list,
> >         >
> >         >         I would like to use PyTrilinos with Anasazi to
> >         compute
> >         >         eigenvectors of
> >         >         large non-symmetric sparse matrices. In order to
> >         familiarize
> >         >         myself with
> >         >         both packages, I tried to build a small example for
> >         computing
> >         >         1-eigenvector of a probability transition matrix for
> >         a random
> >         >         walk on a
> >         >         graph. The solver I tried to use was
> >         BlockKrylovSchur.
> >         >
> >         >         I managed to successfuly build the matrix in
> >         question and even
> >         >         perform
> >         >         some basic operations on it (I checked by printing
> >         the
> >         >         matrix). I also
> >         >         managed to pass the setProblem() sanity test with
> >         the
> >         >         Eigenproblem
> >         >         object. However, when I try to invoke the solver,
> >         the GEES
> >         >         subroutine
> >         >         fails which causes the entire code to fail.
> >         >
> >         >         This is the traceback from Python:
> >         >         ******************
> >         >         Traceback (most recent call last):
> >         >           File "src/anasazitest.py", line 56, in <module>
> >         >             info = mngr.solve()
> >         >           File
> >         >
> >                 "/usr/lib/python2.5/site-packages/PyTrilinos/Anasazi.py", line
> >         >         6456, in solve
> >         >             return
> >         _Anasazi.BlockKrylovSchurSolMgrEpetra_solve(*args)
> >         >
> >         >
> >         >         Throw number = 1
> >         >
> >         >         Throw test that evaluated to true: info != 0
> >         >
> >         >         Anasazi::BlockKrylovSchur::computeSchurForm(): GEES
> >         returned
> >         >         info != 0.
> >         >         *******************
> >         >
> >         >         I am guessing that I am mismatching some options,
> >         possibly
> >         >         related to
> >         >         block size and sizes of auxiliary vectors but I
> >         don't know how
> >         >         to
> >         >         correct the problem. Below is the relevant source
> >         code.
> >         >
> >         >         If anyone could direct me in correcting this
> >         problem, I would
> >         >         be very
> >         >         thankful. Also, if you have any other suggestions,
> >         based on my
> >         >         code,
> >         >         those would also be very much appreciated.
> >         >
> >         >         Best,
> >         >         Marko Budisic
> >         >
> >         >         *********************************
> >         >         # epetra
> >         >         nx = 5
> >         >         numglobals = nx*nx
> >         >         mycomm = Epetra.SerialComm()
> >         >         mymap = Epetra.Map( numglobals, 0, mycomm )
> >         >         W = Epetra.CrsMatrix(Epetra.Copy, mymap, 4)
> >         >
> >         >         # populate matrix
> >         >         W.InsertGlobalValues( 0, [1.,2.,3.,2.], [0,1,2,4] )
> >         >         W.InsertGlobalValues( 1, [1.,2.], [1,2] )
> >         >         W.InsertGlobalValues( 2, [1.,2.], [2,3] )
> >         >         W.InsertGlobalValues( 3, [3.,1.], [0,3] )
> >         >         W.InsertGlobalValues( 4, [2.,1.], [3,4] )
> >         >         W.FillComplete()
> >         >
> >         >         # normalize the matrix
> >         >         d = Epetra.Vector( W.RowMap() )
> >         >         W.InvRowSums(d)
> >         >         W.LeftScale(d)
> >         >
> >         >         print W
> >         >
> >         >         eigs = Anasazi.BasicEigenproblem()
> >         >         eigs.setA( W )
> >         >         eigs.setHermitian(False)
> >         >
> >         >         Nev = 1
> >         >         eigs.setNEV( Nev )
> >         >
> >         >         blocksize = 1
> >         >         storage = Epetra.MultiVector( W.RowMap() ,
> >         blocksize)
> >         >         eigs.setInitVec(storage)
> >         >
> >         >         print "Problem set:", eigs.setProblem()
> >         >
> >         >         myPL = Teuchos.ParameterList()
> >         >         myPL.set("Which", "LM")
> >         >         myPL.set("Block Size",blocksize)
> >         >         myPL.set("Verbosity", Anasazi.Errors |
> >         Anasazi.Warnings | \
> >         >                      Anasazi.FinalSummary )
> >         >
> >         >         mngr = Anasazi.BlockKrylovSchurSolMgr( eigs, myPL )
> >         >         info = mngr.solve()
> >         >         print "Problem solved:", info
> >         >
> >         >         x = eigs.getSolution()
> >         >         print x.Evecs()
> >         >
> >         >         *********************************
> >         >
> >         >
> >         >         _______________________________________________
> >         >         Trilinos-Users mailing list
> >         >         Trilinos-Users at software.sandia.gov
> >         >
> >                 http://software.sandia.gov/mailman/listinfo/trilinos-users
> >         >
> >         >
> >         > _______________________________________________
> >         > Trilinos-Users mailing list
> >         > Trilinos-Users at software.sandia.gov
> >         > http://software.sandia.gov/mailman/listinfo/trilinos-users
> >
> >
> >         _______________________________________________
> >         Trilinos-Users mailing list
> >         Trilinos-Users at software.sandia.gov
> >         http://software.sandia.gov/mailman/listinfo/trilinos-users
> >
> >
> > --------------------------------------
> > Christopher G. Baker, Ph.D.
> > Scalable Algorithms Department (01416)
> > Sandia National Laboratories
> > MS 1320, P.O. Box 5800
> > Albuquerque, NM 87185-1320
> > email: cgbaker at sandia.gov
> > phone: 505-284-6978
> > --------------------------------------
> >
>
>
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/trilinos-users
>

```