Re: [eigen] Predefined SymmetircMatirx, BandedMatrix and PositiveDefiniteMatrix in Eigen

[ Thread Index | Date Index | More Archives ]

On Wed, Jul 20, 2011 at 8:08 AM, Hongyu Miao <jackymiao@xxxxxxxxx> wrote:
> Hi, All
>      I’m wondering why there are not predefined data types for
> SymmetircMatirx, BandedMatrix and PositiveDefiniteMatrix in Eigen. These are
> important in computing and boost::ublas has all such things except for
> PositiveDefiniteMatrix.

But there is a SelfAdjointView class and a .selfadjointView() method
for symmetric matrix manipulations.  Similarly TriangularView and
..triangularView() for triangular matrix manipulations.  ublas uses
packed storage for symmetric and triangular matrices, if I recall
correctly, but Eigen uses the same underlying representation (full
dense or sparse matrices) with additional properties and access
methods.  There are advantages and disadvantages to either approach.

It may be confusing that the term "self-adjoint" is used instead of
"symmetric" but for complex matrices there is a distinction between
symmetric and Hermitian so self-adjoint is a more precise term.

Mail converted by MHonArc 2.6.19+