Did anybody meet a striking disagreement between the output data from
LAPACK routines and other library routines when calculating
eigenvectors of a symmetric matrix?
I took several symmetric 3x3 matrices and used 2+3 subroutines from
LAPACK in various combinations to calculte eigenproblem for a matrix.
They are DSTEQR, DSYTD2 and DSTEDC, DSTEV, DSTEVD.
(1) Running any combination gives the same output data.
Which is great!
(2) But calculated eigenvectors are wrong!
A simple check of their orthoganality properties confirms this.
(3) Moreover, I processed the same matrices with NAG-77 library
and with MUPAD. In both cases I obtained the same correct results
without any problems.
LAPACK I used:
version 3.0 dated by 1999
Some matrix I used:
BRR=RESHAPE( (/10., 2., 3., 2., 10., 4., 3., 4., 10. /), (/ 3,3 /) )
Is anybody able to comment my post?