> HI Martin,
> Yes I do have a matrix close to 30000 variables and 348
> observations. Here, by variables I mean are the points on a rectangular
> grid and observations are the values at each grid measured every month for
> almost 3 decades. If the matrix contains no information how to avoid this
> and compute.
> Thanks and Regards
> nuncio
>
> On Wed, Apr 6, 2011 at 7:52 PM, Martin Helm <
address@hidden> wrote:
> > Am Mittwoch, den 06.04.2011, 10:11 +0530 schrieb nuncio m:
> > HI martin,
> >
> > > Thanks, but I need to find the covariance matrix
> > >
> > > separately. Because as a next step I need to find the svd of coupled
> > > fields. for example atmospheric pressure and temperature. For a
> > > single matrix I can find eigen vectors using SVD but I understand
> > > this
> > > is not possible without computing the covariance matrix for two
> > > separate matrices.
> > > nuncio
> >
> > Dear Nuncio,
> >
> > sorry that I insist so much. I do of course not know your problem at
> > hand, but have my doubts about the matrix you compute (the product which
> > blows up to about 30000x30000). This matrix contains essentially no
> > information, more than 29000 of its eigen values are exact zero, more
> > than 29000 of its eigen vectors are simply a representation of the null
> > space since it has a maximal rank of 348.
> >
> > Do you really have nearly 30000 variables and only 348 measurements?
> >
> > - Martin