Confusion about output

[Antonio Medrano]

In the Matlab code, the output of the program is relres

relres = sqrt(rtr) / normb, where normb = norm(b).

In the MPI code, the output is the value norm. What exactly is this norm
value of? Is it norm(r)? Or should we output the equivalent to relres
from the matlab code?

-Antonio

[Matt Weiden]

In the email the professor sent to me he asks for the "norm of the residual". So in the notation from the lecture slides I believe that would be norm(r). Does that sound right?

Matt

[Rachel L. Moore]

The output is norm(r) but for calculating relres you use the norm(b),
which never changes. If you use norm(r) for relres I don't think it
will converge.

Rachel

Quoting Matt Weiden <wei...@cs.ucsb.edu>:

--
Rachel Moore
Department of Economics
University of California Santa Barbara

[John Gilbert]

Sorry for the confusion here.  The convergence test for CG is usually to stop when the residual r = b-Ax is small compared to b, the right-hand side.  That's measured by the "relative residual", which is norm(r)/norm(b).  As Rachel says, norm(b) doesn't change during the CG iteration.  We probably should have made the harness code print norm(r)/norm(b).

Thanks,

- John