<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    <div class="moz-cite-prefix">Matt<br>
      <br>
      Sorry about that (I always forget it). The output for the smallest
      problem is now attached (see log.txt). I am also attaching some
      results that compare results obtained using FS/LSC and the direct
      solver (MUMPS), again for the smallest problem. The difference, as
      you can see is insignificant O(1E-6).<br>
      <br>
      Also, I did pass 'upper' and 'full' to
      '-pc_fieldsplit_schur_factorization_type' but the iteration count
      doesn't improve (in fact, it increases slightly). The attached log
      is with 'upper'.<br>
      <br>
      Regards,<br>
      <br>
      Tabrez<br>
      <br>
      On 10/23/2014 07:46 AM, Matthew Knepley wrote:<br>
    </div>
    <blockquote
cite="mid:CAMYG4G=3EKCZMjnO9ea=d4VPNjZaZGOw5_7m8B-d+Fp-ae52gw@mail.gmail.com"
      type="cite">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">On Thu, Oct 23, 2014 at 7:20 AM,
            Tabrez Ali <span dir="ltr"><<a moz-do-not-send="true"
                href="mailto:stali@geology.wisc.edu" target="_blank">stali@geology.wisc.edu</a>></span>
            wrote:<br>
            <blockquote class="gmail_quote" style="margin:0 0 0
              .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div bgcolor="#FFFFFF" text="#000000"> Hello<br>
                <br>
                I am using the following options (below) for solving
                linear elasticity/poroelasticity problems involving slip
                between two surfaces involving non-trivial geometries,
                i.e., elements with high aspect ratios, large contrasts
                in material properties etc. The constraints are imposed
                using Lagrange Multipliers. <br>
                <br>
                A picture (shows displacement magnitude) is attached.
                The boundary nodes, i.e., the base and the four side are
                pinned.<br>
                <br>
                The following options appear to work well for the saddle
                point problem:<br>
                <br>
                <tt>-pc_type fieldsplit -pc_fieldsplit_type schur
                  -pc_fieldsplit_detect_saddle_point
                  -fieldsplit_0_pc_type gamg -fieldsplit_0_ksp_type
                  preonly -fieldsplit_1_pc_type lsc
                  -fieldsplit_1_ksp_type preonly
                  -pc_fieldsplit_schur_fact_type lower -ksp_monitor</tt><br>
                <br>
                However, the number of iterations keep on increasing
                with the problems size (see attached plot), e.g.,<br>
                <br>
                <tt>120K Tets        <b>507</b> Iterations (KSP
                  Residual norm 8.827362494659e-05)</tt><tt> in  17 secs
                  on   3 cores<br>
                </tt><tt>1 Million Tets  <b>1374</b> Iterations (KSP
                  Residual norm 7.164704416296e-05)</tt><tt> in 117 secs
                  on  20 cores<br>
                </tt><tt>8 Million Tets  <b>2495</b> Iterations (KSP
                  Residual norm 9.101247550026e-05) in 225 secs on 160
                  cores</tt><br>
                <br>
                So what other options should I try to improve solver
                performance? Any tips/insights would be appreciated as
                preconditioning is black magic to me.<br>
              </div>
            </blockquote>
            <div><br>
            </div>
            <div>For reports, always run with </div>
            <div><br>
            </div>
            <div>  -ksp_view -ksp_monitor_true_residual
              -ksp_converged_reason</div>
            <div><br>
            </div>
            <div>so that we can see exactly what you used.</div>
            <div><br>
            </div>
            <div>I believe the default is a diagonal factorization.
              Since your outer iterates are increasing, I would
              strengthen this</div>
            <div>to either upper or full</div>
            <div><br>
            </div>
            <div>  -pc_fieldsplit_schur_factorization_type <upper,
              full></div>
            <div><br>
            </div>
            <div>  Thanks,</div>
            <div><br>
            </div>
            <div>      Matt </div>
            <div> </div>
            <blockquote class="gmail_quote" style="margin:0 0 0
              .8ex;border-left:1px #ccc solid;padding-left:1ex">
              <div bgcolor="#FFFFFF" text="#000000"> Thanks in advance.<span
                  class="HOEnZb"><font color="#888888"><br>
                    <br>
                    Tabrez<br>
                  </font></span></div>
            </blockquote>
          </div>
          <br>
          <br clear="all">
          <div><br>
          </div>
          -- <br>
          What most experimenters take for granted before they begin
          their experiments is infinitely more interesting than any
          results to which their experiments lead.<br>
          -- Norbert Wiener </div>
      </div>
    </blockquote>
    <br>
  </body>
</html>