[Trilinos-Users] Parallel Epetra_LinearProblem

[Williams, Alan B]

John,
I'm not exactly sure what the problem is here, but...

It is generally difficult to come up with the correct column-map in parallel, you usually want to just construct the graph and/or matrix with a row-map, and let it create its own column-map during the FillComplete method.

For a description of the four maps and how they relate to each other, take a look at the doxygen documentation for the Epetra_VbrMatrix (go to trilinos.sandia.gov, go to the epetra package, dive, dive...)

Alan


> -----Original Message-----
> From: trilinos-users-bounces at software.sandia.gov [mailto:trilinos-
> users-bounces at software.sandia.gov] On Behalf Of John Mitchell
> Sent: Thursday, June 10, 2010 2:29 PM
> To: trilinos-users at software.sandia.gov
> Subject: Re: [Trilinos-Users] Parallel Epetra_LinearProblem
> 
> After further investigation -- I have uncovered a piece of evidence
> that
> I don't understand.
> 
> I have explicitly constructed an Epetra_CrsGraph using a different
> rowMap and columnMap:
> Epetra_CrsGraph *graph = new
> Epetra_CrsGraph(Copy,rowMap,colMap,numCols.get(),true);
> 
> Later, I constructed a Epetra_FEVbrMatrix
>  shared_ptr<Epetra_FEVbrMatrix> mPtr(new
> Epetra_FEVbrMatrix(Copy,*(graph)));
> 
> I then used some debug commands to print out the underlying data that
> each map stores -- this is used in the "CheckInput" of a
> Epetra_LinearProblem:
>     cout << "computeJacobian:: mPtr->OperatorDomainMap().DataPtr() = "
> << mPtr->OperatorDomainMap().DataPtr() << endl;
>     cout << "computeJacobian:: rowMap.DataPtr() = " << rowMap.DataPtr()
> << endl;
>     cout << "computeJacobian::mPtr->OperatorRangeMap().DataPtr() = " <<
> mPtr->OperatorRangeMap().DataPtr() << endl;
>     cout << "computeJacobian:: colMap.DataPtr() = " << colMap.DataPtr()
> << endl;
> 
> Results of the above print out don't make sense to me?  Notice how the
> Operator range and domain maps are the same?  Also notice that neither
> one matches the rowMap or colMap respectively?  I'm not quite sure how
> the operator domain and range maps are the same since I explicitly
> constructed the Operator with a different row and col maps.
> 
> computeJacobian:: mPtr->OperatorDomainMap().DataPtr() = 0x1b123cf0
> computeJacobian:: rowMap.DataPtr() = 0x1b11ee40
> computeJacobian::mPtr->OperatorRangeMap().DataPtr() = 0x1b123cf0
> computeJacobian:: colMap.DataPtr() = 0x1b123460
> 
> Any comments or suggestions would be appreciated.
> Thanks,
> John
> 
> John Mitchell wrote:
> > Hi,
> >
> > I am having trouble getting past the "CheckInput" stage for a
> > Epetra_LinearProblem.  In serial the problem runs fine, but in
> parallel
> > I fail 'CheckInput'.
> >
> > I successfully constructed and filled a graph checking all error
> flags
> > -- no failures or warnings.
> >
> > Similarly, I successfully constructed an Epetra_RowMatrix with the
> above
> > graph and then filled it checking all error flags -- no failures or
> > warnings.
> >
> > The last step is to create an Epetra_LinearProblem.
> >
> > Epetra_RowMatrix *mPtr = ...
> > Epetra_LinearProblem linProblem;
> > linProblem.SetOperator(mPtr);
> > linProblem.AssertSymmetric();
> >
> > When I constructed both the Epetra_CrsGraph and Epetra_FEVbrMatrix I
> > used the following maps.
> >     const Epetra_BlockMap& rowMap  = jacobian->getRowMap();
> >     const Epetra_BlockMap& colMap  = jacobian->getColMap();
> >
> > Now, I created the RHS and LHS Epetra_Vectors : (this is simplified
> to
> > emphasize the maps I used)
> >     double *u = ...;
> >     double *f = ...;
> >     Epetra_Vector rhs(View,rowMap,f);
> >     Epetra_Vector lhs(View,colMap,u);
> >     linProblem.SetRHS(&rhs);
> >     linProblem.SetLHS(&lhs);
> >     BOOST_CHECK(0==linProblem.CheckInput());
> >
> > I am failing the last check at the bottom with the error flag = -5?
> >
> > I thought that I was very careful to use the proper maps.  When I run
> > this in serial it runs correctly.  But when I run it in parallel it
> > fails the last error flag check?
> >
> > Any thoughts/suggestions would be much appreciated.
> >
> > Thanks,
> > John
> >
> >
> >
> >
> > _______________________________________________
> > 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