<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">On 25/10/13 18:59, Matthew Knepley
      wrote:<br>
    </div>
    <blockquote
cite="mid:CAMYG4GntLCk9WCOE84tfMbPFG=xGc03DY5oZvWShvOm1grAuwQ@mail.gmail.com"
      type="cite">
      <div dir="ltr">On Fri, Oct 25, 2013 at 11:33 AM, Torquil Macdonald
        Sørensen <span dir="ltr"><<a moz-do-not-send="true"
            href="mailto:torquil@gmail.com" target="_blank">torquil@gmail.com</a>></span>
        wrote:<br>
        <div class="gmail_extra">
          <div class="gmail_quote">
            <blockquote class="gmail_quote" style="margin:0px 0px 0px
0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
              <div dir="ltr">
                <div>
                  <div>
                    <div>
                      <div>Hi!<br>
                        <br>
                      </div>
                      I'm using PETSc 3.4.3, with only default runtime
                      settings, to solve Ax=b, for complex numbers (I
                      have used --with-scalar-type=complex).<br>
                      <br>
                    </div>
                    After reading sections 4.3.1, 4.3.2, 4.3.3 in the
                    manual, I decided to investigate the convergence in
                    my program, since I'm having some problems when the
                    system is larger. So, I'm running it with the
                    options<br>
                    <br>
                    -ksp_monitor_true_residual -ksp_converged_reason
                    -ksp_view<br>
                    <br>
                  </div>
                  The output shows that the relative tolerance required
                  for convergence is 1e-5, but none of the reported
                  norms are that small. Despite this, convergence is
                  reported (CONVERGED_RTOL).<br>
                  <br>
                </div>
                <div>I though that CONVERGED_RTOL should indicate that
                  ||r||_2/||b||_2 < 1e-5, where r is the
                  preconditioned residual?<br>
                </div>
                <div><br>
                </div>
                <div>Here is the output of my program (which includes
                  some code to compute the l_2 norm of b, just before
                  KSPSolve is run):<br>
                  <br>
*************************************************************<br>
                  Norm of b: 0.179829<br>
                    0 KSP preconditioned resid norm 2.508789303280e+04
                  true resid norm 1.798290175843e-01 ||r(i)||/||b||
                  1.000000000000e+00<br>
                    1 KSP preconditioned resid norm 1.856991132478e+00
                  true resid norm 3.186462510386e-01 ||r(i)||/||b||
                  1.771940120227e+00<br>
                    2 KSP preconditioned resid norm 1.704678606994e-01
                  true resid norm 4.004071321891e-02 ||r(i)||/||b||
                  2.226599119363e-01<br>
                </div>
              </div>
            </blockquote>
            <div><br>
            </div>
            <div>So the relative preconditioned residual is:</div>
            <div>
              <br>
            </div>
            <div>  1.7e-01 / 2.5e+04 < 1.0e-5</div>
            <div><br>
            </div>
            <div>while the relative true residual is 2.2e-1. This
              happens because you are using ILU(0). It can be really</div>
            <div>crappy, and in this case the preconditioner is very
              ill-conditioned. I can't imagine a problem where I</div>
            <div>would recommend using ILU, but its our default because
              its the only black-box PC that exists.</div>
            <div><br>
            </div>
            <div>  Thanks</div>
            <div><br>
            </div>
            <div>     Matt</div>
            <div> </div>
            <br>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
    From your explanation I can see that I had misunderstood the meaning
    of "relative preconditioned residual norm", but now I understand.
    Thanks for clearing it up!<br>
    <br>
    Best regards<br>
    Torquil Sørensen<br>
  </body>
</html>