




Fortran Programming Language









Arpack usage
Hello, I am now involved in using arpack to solve a large complex matrix's eigenvalues and corresponding eigenvectors. The solver I used is arpack, it's really great, fast and convenient. I compare the eigenvalues and eigenvectors from arpack solver and matlab 'eigs' function (though I know it also uses arpack), however, I found a big mismatch on eigenvectors (eigenvalue is OK, they match!) Can anyone who have used arpack tell me why such big mismatch exist? Thanks a lot, lorin
On May 14, 9:29 am, lorin <wwang@gmail.com> wrote: > Hello, > I am now involved in using arpack to solve a large complex matrix's > eigenvalues and corresponding eigenvectors. The solver I used is > arpack, it's really great, fast and convenient. > I compare the eigenvalues and eigenvectors from arpack solver and > matlab 'eigs' function (though I know it also uses arpack), however, I > found a big mismatch on eigenvectors (eigenvalue is OK, they match!) > Can anyone who have used arpack tell me why such big mismatch exist?
If X is an eigenvector, so c*X (a multiple of X), and arpack and Matlab may use different conventions in scaling the eigenvectors. Are the arpack eigenvectors proportional to those of Matlab?
On 14 mei, 15:29, lorin <wwang@gmail.com> wrote: > Hello, > I am now involved in using arpack to solve a large complex matrix's > eigenvalues and corresponding eigenvectors. The solver I used is > arpack, it's really great, fast and convenient. > I compare the eigenvalues and eigenvectors from arpack solver and > matlab 'eigs' function (though I know it also uses arpack), however, I > found a big mismatch on eigenvectors (eigenvalue is OK, they match!) > Can anyone who have used arpack tell me why such big mismatch exist? > Thanks a lot, > lorin
Perhaps you have eigenvalues that (nearly) coincide? Then you could get any linear combination of the eigenvectors. That is one possible explanation. Another is: have you normalised the vectors before comparing them? There could be a (complex) factor between them. Regards, Arjen
On May 14, 9:42 pm, Arjen Markus <arjen.mar@wldelft.nl> wrote:
> On 14 mei, 15:29, lorin <wwang @gmail.com> wrote: > > Hello, > > I am now involved in using arpack to solve a large complex matrix's > > eigenvalues and corresponding eigenvectors. The solver I used is > > arpack, it's really great, fast and convenient. > > I compare the eigenvalues and eigenvectors from arpack solver and > > matlab 'eigs' function (though I know it also uses arpack), however, I > > found a big mismatch on eigenvectors (eigenvalue is OK, they match!) > > Can anyone who have used arpack tell me why such big mismatch exist? > > Thanks a lot, > > lorin > Perhaps you have eigenvalues that (nearly) coincide? Then you could > get any > linear combination of the eigenvectors. That is one possible > explanation. > Another is: have you normalised the vectors before comparing them? > There > could be a (complex) factor between them. > Regards, > Arjen
Thanks, Arjen and Beliavsky... I have recompare the eigenvectors from Matlab and ARPACK. There is a factor between them, for different matrix, the factor is not the same, but all factors' absolute magnitude is unit length. I try to find out why such factor exists, but it seems difficult...





