Using the four figure decimals that you quote I get
residuals (r = A*x - lambda*B*x) such that
norm(r1) = 1.5921e-04, norm(r2) = 6.0842e-05.
Since norm(A) = 1.2994 and norm(B) = 7.9874, these residuals seem
The eigenvectors produced by DSPGVX are normalized so that
norm(x'*B*x) = 1.
[ ... ]
DSPGVX will have computed the solution in double precision,
but the web site has rounded the solution to four decimal figures for
convenience of printed output. If I take the solution direct from
DSPGVX then on my machine I get residuals
norm(r1) = 5.4177e-16, norm(r2) = 5.3663e-16,
so DSPGVX is indeed computing satisfactory results.
Users browsing this forum: Google [Bot] and 2 guests