dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, Apr 04, 2011 at 08:46:48PM +0100, Garth N. Wells wrote:
>
>
> On 04/04/11 20:20, Johan Hake wrote:
> > On Monday April 4 2011 12:13:22 Anders Logg wrote:
> >> On Mon, Apr 04, 2011 at 08:10:22PM +0100, Garth N. Wells wrote:
> >>> On 04/04/11 20:00, Johan Hake wrote:
> >>>> On Monday April 4 2011 11:55:05 Garth N. Wells wrote:
> >>>>> On 04/04/11 19:51, Johan Hake wrote:
> >>>>>> Hello!
> >>>>>>
> >>>>>> I would very much like to have something like:
> >>>>>>   // Set the diagonal of a matrix
> >>>>>>   GenericMatrix::set_diagonal(const GenericVector& vec);
> >>>>>>
> >>>>>>   // Scale Matrix with a diagonal matrix (represented as a Vector)
> >>>>>>   GenericMatrix::scale(const GenericVector& vec);
> >>>>>>   (maybe use operator*=)
> >>>>>>
> >>>>>> added to the GenericMatrix interface.
> >>>>>>
> >>>>>> The corresponding backend specific methods would be:
> >>>>>>   // PETSc
> >>>>>>   MatDiagonalSet
> >>>>>>   MatDiagonalScale
> >>>>>>
> >>>>>>   // Epetra
> >>>>>>   ReplaceDiagonalValues
> >>>>>>   RightScale
> >>>>>>
> >>>>>>   // MTL4/uBLAS find out later...
> >>>>>>
> >>>>>> which should ensure the methods works in parallel too.
> >>>>>>
> >>>>>> Any objections or advices in respect to implement these?
> >>>>>
> >>>>> Fine with me, but could GenericMatrix::scale have a better name? It's
> >>>>> not clear to me what it would do. Scale the diagonal, or scale the
> >>>>> rows or the columns?
> >>>>
> >>>> It basically doing:
> >>>>   A*=D
> >>>>
> >>>> where D is a diagonal matrix given by the a vector. And I guess this
> >>>> would then correspond to a column scale.
> >>>
> >>> OK. It might get complicated, but we could have a
> >>>
> >>>     (Generic)DiagonalMatrix
> >>>
> >>> class?
> >
> > Hmm, letting it inherit GenericVector? This path would open up a box of things
> > we have to consider... Maye fire up a Blueprint and add other thoughts on
> > that.
> >
> >>> Otherwise, GenericMatrix::right_scale is probably an OK name.
> >
> > Ok.
> >
> >> I think scale_columns is more informative, if that is what it actually
> >> does.
> >>
> >> (Is it implied by A *= D that it corresponds to A = A * D or can it be
> >> A = D * A?)
> >
> > Yes, no.
> >
> > The first one is called RightScale and the second (I guess) is called
> > LeftScale in Epetra. In PETSc it is put into the same function.
> >
> > I could add right_scale and left_scale, or row_scale and column_scale. No big
> > deal from me.
> >
>
> 'left' and 'right' is consistent with preconditioner terminology, and
> with the Trilinos and PETSc naming.

Yes perhaps, but I still thing scale_rows / scale_columns are more
informative.

--
Anders


> Garth
>
> > Johan
> >
> >>
> >>> Garth
> >>>
> >>>> Johan
> >>>>
> >>>>> Garth
> >>>>>
> >>>>>> Best regards,
> >>>>>>
> >>>>>> Johan
> >>>>>>
> >>>>>> _______________________________________________
> >>>>>> Mailing list: https://launchpad.net/~dolfin
> >>>>>> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> >>>>>> Unsubscribe : https://launchpad.net/~dolfin
> >>>>>> More help   : https://help.launchpad.net/ListHelp
> >>>>>
> >>>>> _______________________________________________
> >>>>> Mailing list: https://launchpad.net/~dolfin
> >>>>> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> >>>>> Unsubscribe : https://launchpad.net/~dolfin
> >>>>> More help   : https://help.launchpad.net/ListHelp
> >>>
> >>> _______________________________________________
> >>> Mailing list: https://launchpad.net/~dolfin
> >>> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> >>> Unsubscribe : https://launchpad.net/~dolfin
> >>> More help   : https://help.launchpad.net/ListHelp