<html>
  <head>
    <meta content="text/html; charset=UTF-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    This is what I get for 3.5.0:<br>
    <br>
    anton@anton ~/LIB/petsc-3.5.0-deb/src/snes/examples/tutorials $
    ./ex19 -bf_pc_type fieldsplit -bf_snes_view<br>
    lid velocity = 0.0625, prandtl # = 1, grashof # = 1<br>
    Number of SNES iterations = 2<br>
    WARNING! There are options you set that were not used!<br>
    WARNING! could be spelling mistake, etc!<br>
    Option left: name:-bf_pc_type value: fieldsplit<br>
    Option left: name:-bf_snes_view (no value)<br>
    <br>
    It seems like it was already corrected  between 3.5.0 & 3.5.2<br>
    <br>
    <div class="moz-cite-prefix">On 09/26/2014 04:26 PM, Matthew Knepley
      wrote:<br>
    </div>
    <blockquote
cite="mid:CAMYG4GntYGZvi2GE3hdNNYO37CdKwTbMjhJG9wtOMXo0z2KLvQ@mail.gmail.com"
      type="cite">
      <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
      <div dir="ltr">Here is the result for 3.5.2, which looks right to
        me:
        <div><br>
        </div>
        <div>
          <div>(v3.5.2)
            *:/PETSc3/petsc/release-petsc-3.5.1/src/snes/examples/tutorials$
            ./ex19 -bf_pc_type fieldsplit -bf_snes_view
            -bf_pc_fieldsplit_type schur -bf_pc_fieldsplit_0_fields
            0,1,2 -bf_pc_fieldsplit_1_fields 3
            -bf_pc_fieldsplit_schur_factorization_type upper</div>
          <div>lid velocity = 0.0625, prandtl # = 1, grashof # = 1</div>
          <div>SNES Object:(bf_) 1 MPI processes</div>
          <div>  type: newtonls</div>
          <div>  maximum iterations=50, maximum function
            evaluations=10000</div>
          <div>  tolerances: relative=1e-08, absolute=1e-50,
            solution=1e-08</div>
          <div>  total number of linear solver iterations=4</div>
          <div>  total number of function evaluations=3</div>
          <div>  SNESLineSearch Object:  (bf_)   1 MPI processes</div>
          <div>    type: bt</div>
          <div>      interpolation: cubic</div>
          <div>      alpha=1.000000e-04</div>
          <div>    maxstep=1.000000e+08, minlambda=1.000000e-12</div>
          <div>    tolerances: relative=1.000000e-08,
            absolute=1.000000e-15, lambda=1.000000e-08</div>
          <div>    maximum iterations=40</div>
          <div>  KSP Object:  (bf_)   1 MPI processes</div>
          <div>    type: gmres</div>
          <div>      GMRES: restart=30, using Classical (unmodified)
            Gram-Schmidt Orthogonalization with no iterative refinement</div>
          <div>      GMRES: happy breakdown tolerance 1e-30</div>
          <div>    maximum iterations=10000, initial guess is zero</div>
          <div>    tolerances:  relative=1e-05, absolute=1e-50,
            divergence=10000</div>
          <div>    left preconditioning</div>
          <div>    using PRECONDITIONED norm type for convergence test</div>
          <div>  PC Object:  (bf_)   1 MPI processes</div>
          <div>    type: fieldsplit</div>
          <div>      FieldSplit with Schur preconditioner, factorization
            UPPER</div>
          <div>      Preconditioner for the Schur complement formed from
            A11</div>
          <div>      Split info:</div>
          <div>      Split number 0 Defined by IS</div>
          <div>      Split number 1 Defined by IS</div>
          <div>      KSP solver for A00 block</div>
          <div>        KSP Object:        (bf_fieldsplit_0_)         1
            MPI processes</div>
          <div>          type: gmres</div>
          <div>            GMRES: restart=30, using Classical
            (unmodified) Gram-Schmidt Orthogonalization with no
            iterative refinement</div>
          <div>            GMRES: happy breakdown tolerance 1e-30</div>
          <div>          maximum iterations=10000, initial guess is zero</div>
          <div>          tolerances:  relative=1e-05, absolute=1e-50,
            divergence=10000</div>
          <div>          left preconditioning</div>
          <div>          using PRECONDITIONED norm type for convergence
            test</div>
          <div>        PC Object:        (bf_fieldsplit_0_)         1
            MPI processes</div>
          <div>          type: ilu</div>
          <div>            ILU: out-of-place factorization</div>
          <div>            0 levels of fill</div>
          <div>            tolerance for zero pivot 2.22045e-14</div>
          <div>            using diagonal shift on blocks to prevent
            zero pivot [INBLOCKS]</div>
          <div>            matrix ordering: natural</div>
          <div>            factor fill ratio given 1, needed 1</div>
          <div>              Factored matrix follows:</div>
          <div>                Mat Object:                 1 MPI
            processes</div>
          <div>                  type: seqaij</div>
          <div>                  rows=48, cols=48</div>
          <div>                  package used to perform factorization:
            petsc</div>
          <div>                  total: nonzeros=576, allocated
            nonzeros=576</div>
          <div>                  total number of mallocs used during
            MatSetValues calls =0</div>
          <div>                    using I-node routines: found 16
            nodes, limit used is 5</div>
          <div>          linear system matrix = precond matrix:</div>
          <div>          Mat Object:          (bf_fieldsplit_0_)        
              1 MPI processes</div>
          <div>            type: seqaij</div>
          <div>            rows=48, cols=48</div>
          <div>            total: nonzeros=576, allocated nonzeros=576</div>
          <div>            total number of mallocs used during
            MatSetValues calls =0</div>
          <div>              using I-node routines: found 16 nodes,
            limit used is 5</div>
          <div>      KSP solver for S = A11 - A10 inv(A00) A01 </div>
          <div>        KSP Object:        (bf_fieldsplit_temperature_)  
                  1 MPI processes</div>
          <div>          type: gmres</div>
          <div>            GMRES: restart=30, using Classical
            (unmodified) Gram-Schmidt Orthogonalization with no
            iterative refinement</div>
          <div>            GMRES: happy breakdown tolerance 1e-30</div>
          <div>          maximum iterations=10000, initial guess is zero</div>
          <div>          tolerances:  relative=1e-05, absolute=1e-50,
            divergence=10000</div>
          <div>          left preconditioning</div>
          <div>          using PRECONDITIONED norm type for convergence
            test</div>
          <div>        PC Object:        (bf_fieldsplit_temperature_)  
                  1 MPI processes</div>
          <div>          type: ilu</div>
          <div>            ILU: out-of-place factorization</div>
          <div>            0 levels of fill</div>
          <div>            tolerance for zero pivot 2.22045e-14</div>
          <div>            using diagonal shift on blocks to prevent
            zero pivot [INBLOCKS]</div>
          <div>            matrix ordering: natural</div>
          <div>            factor fill ratio given 1, needed 1</div>
          <div>              Factored matrix follows:</div>
          <div>                Mat Object:                 1 MPI
            processes</div>
          <div>                  type: seqaij</div>
          <div>                  rows=16, cols=16</div>
          <div>                  package used to perform factorization:
            petsc</div>
          <div>                  total: nonzeros=64, allocated
            nonzeros=64</div>
          <div>                  total number of mallocs used during
            MatSetValues calls =0</div>
          <div>                    not using I-node routines</div>
          <div>          linear system matrix followed by preconditioner
            matrix:</div>
          <div>          Mat Object:        
             (bf_fieldsplit_temperature_)           1 MPI processes</div>
          <div>            type: schurcomplement</div>
          <div>            rows=16, cols=16</div>
          <div>              Schur complement A11 - A10 inv(A00) A01</div>
          <div>              A11</div>
          <div>                Mat Object:              
             (bf_fieldsplit_temperature_)                 1 MPI
            processes</div>
          <div>                  type: seqaij</div>
          <div>                  rows=16, cols=16</div>
          <div>                  total: nonzeros=64, allocated
            nonzeros=64</div>
          <div>                  total number of mallocs used during
            MatSetValues calls =0</div>
          <div>                    not using I-node routines</div>
          <div>              A10</div>
          <div>                Mat Object:                 1 MPI
            processes</div>
          <div>                  type: seqaij</div>
          <div>                  rows=16, cols=48</div>
          <div>                  total: nonzeros=192, allocated
            nonzeros=192</div>
          <div>                  total number of mallocs used during
            MatSetValues calls =0</div>
          <div>                    not using I-node routines</div>
          <div>              KSP of A00</div>
          <div>                KSP Object:              
             (bf_fieldsplit_0_)                 1 MPI processes</div>
          <div>                  type: gmres</div>
          <div>                    GMRES: restart=30, using Classical
            (unmodified) Gram-Schmidt Orthogonalization with no
            iterative refinement</div>
          <div>                    GMRES: happy breakdown tolerance
            1e-30</div>
          <div>                  maximum iterations=10000, initial guess
            is zero</div>
          <div>                  tolerances:  relative=1e-05,
            absolute=1e-50, divergence=10000</div>
          <div>                  left preconditioning</div>
          <div>                  using PRECONDITIONED norm type for
            convergence test</div>
          <div>                PC Object:              
             (bf_fieldsplit_0_)                 1 MPI processes</div>
          <div>                  type: ilu</div>
          <div>                    ILU: out-of-place factorization</div>
          <div>                    0 levels of fill</div>
          <div>                    tolerance for zero pivot 2.22045e-14</div>
          <div>                    using diagonal shift on blocks to
            prevent zero pivot [INBLOCKS]</div>
          <div>                    matrix ordering: natural</div>
          <div>                    factor fill ratio given 1, needed 1</div>
          <div>                      Factored matrix follows:</div>
          <div>                        Mat Object:                      
              1 MPI processes</div>
          <div>                          type: seqaij</div>
          <div>                          rows=48, cols=48</div>
          <div>                          package used to perform
            factorization: petsc</div>
          <div>                          total: nonzeros=576, allocated
            nonzeros=576</div>
          <div>                          total number of mallocs used
            during MatSetValues calls =0</div>
          <div>                            using I-node routines: found
            16 nodes, limit used is 5</div>
          <div>                  linear system matrix = precond matrix:</div>
          <div>                  Mat Object:                
             (bf_fieldsplit_0_)                   1 MPI processes</div>
          <div>                    type: seqaij</div>
          <div>                    rows=48, cols=48</div>
          <div>                    total: nonzeros=576, allocated
            nonzeros=576</div>
          <div>                    total number of mallocs used during
            MatSetValues calls =0</div>
          <div>                      using I-node routines: found 16
            nodes, limit used is 5</div>
          <div>              A01</div>
          <div>                Mat Object:                 1 MPI
            processes</div>
          <div>                  type: seqaij</div>
          <div>                  rows=48, cols=16</div>
          <div>                  total: nonzeros=192, allocated
            nonzeros=192</div>
          <div>                  total number of mallocs used during
            MatSetValues calls =0</div>
          <div>                    using I-node routines: found 16
            nodes, limit used is 5</div>
          <div>          Mat Object:        
             (bf_fieldsplit_temperature_)           1 MPI processes</div>
          <div>            type: seqaij</div>
          <div>            rows=16, cols=16</div>
          <div>            total: nonzeros=64, allocated nonzeros=64</div>
          <div>            total number of mallocs used during
            MatSetValues calls =0</div>
          <div>              not using I-node routines</div>
          <div>    linear system matrix = precond matrix:</div>
          <div>    Mat Object:     1 MPI processes</div>
          <div>      type: seqaij</div>
          <div>      rows=64, cols=64, bs=4</div>
          <div>      total: nonzeros=1024, allocated nonzeros=1024</div>
          <div>      total number of mallocs used during MatSetValues
            calls =0</div>
          <div>        using I-node routines: found 16 nodes, limit used
            is 5</div>
          <div>Number of SNES iterations = 2</div>
        </div>
        <div><br>
        </div>
        <div>  Thanks,</div>
        <div><br>
        </div>
        <div>     Matt</div>
        <div><br>
        </div>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Fri, Sep 26, 2014 at 8:52 AM,
          Matthew Knepley <span dir="ltr"><<a moz-do-not-send="true"
              href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div dir="ltr">
              <div class="gmail_extra">
                <div class="gmail_quote">
                  <div>
                    <div class="h5">On Fri, Sep 26, 2014 at 7:29 AM,
                      anton <span dir="ltr"><<a
                          moz-do-not-send="true"
                          href="mailto:popov@uni-mainz.de"
                          target="_blank">popov@uni-mainz.de</a>></span>
                      wrote:<br>
                      <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">Create
                        preconditioner:<br>
                        <br>
                        PCCreate(PETSC_COMM_WORLD, &pc);<br>
                        PCSetOptionsPrefix(pc, "bf_");<br>
                        PCSetFromOptions(pc);<br>
                        <br>
                        Define fieldsplit options:<br>
                        <br>
                        -bf_pc_type fieldsplit<br>
                        -bf_pc_fieldsplit_type SCHUR<br>
                        -bf_pc_fieldsplit_schur_factorization_type UPPER<br>
                        <br>
                        Works OK.<br>
                        <br>
                        Set options for the first field solver:<br>
                        <br>
                        -bf_fieldsplit_0_ksp_type preonly<br>
                        -bf_fieldsplit_0_pc_type lu<br>
                        <br>
                        Doesn't work (ignored), because "bf_" prefix
                        isn't pass to inner solver ksp (checked in the
                        debugger).<br>
                        <br>
                        Indeed, the following works:<br>
                        <br>
                        -fieldsplit_0_ksp_type preonly<br>
                        -fieldsplit_0_pc_type lu<br>
                        <br>
                        Observed with 3.5 but not with 3.4<br>
                      </blockquote>
                      <div><br>
                      </div>
                    </div>
                  </div>
                  <div>I just tried this with master on SNES ex19, and
                    got the correct result:</div>
                  <div><br>
                  </div>
                  <div>
                    <div>knepley/feature-parallel-partition
                      *$:/PETSc3/petsc/petsc-dev/src/snes/examples/tutorials$
                      ./ex19 -bf_pc_type fieldsplit -bf_snes_view</div>
                    <div>./ex19 -bf_pc_type fieldsplit -bf_snes_view</div>
                    <div>lid velocity = 0.0625, prandtl # = 1, grashof #
                      = 1</div>
                    <div>SNES Object:(bf_) 1 MPI processes</div>
                    <div>  type: newtonls</div>
                    <div>  maximum iterations=50, maximum function
                      evaluations=10000</div>
                    <div>  tolerances: relative=1e-08, absolute=1e-50,
                      solution=1e-08</div>
                    <div>  total number of linear solver iterations=13</div>
                    <div>  total number of function evaluations=3</div>
                    <div>  SNESLineSearch Object:  (bf_)   1 MPI
                      processes</div>
                    <div>    type: bt</div>
                    <div>      interpolation: cubic</div>
                    <div>      alpha=1.000000e-04</div>
                    <div>    maxstep=1.000000e+08,
                      minlambda=1.000000e-12</div>
                    <div>    tolerances: relative=1.000000e-08,
                      absolute=1.000000e-15, lambda=1.000000e-08</div>
                    <div>    maximum iterations=40</div>
                    <div>  KSP Object:  (bf_)   1 MPI processes</div>
                    <div>    type: gmres</div>
                    <div>      GMRES: restart=30, using Classical
                      (unmodified) Gram-Schmidt Orthogonalization with
                      no iterative refinement</div>
                    <div>      GMRES: happy breakdown tolerance 1e-30</div>
                    <div>    maximum iterations=10000, initial guess is
                      zero</div>
                    <div>    tolerances:  relative=1e-05,
                      absolute=1e-50, divergence=10000</div>
                    <div>    left preconditioning</div>
                    <div>    using PRECONDITIONED norm type for
                      convergence test</div>
                    <div>  PC Object:  (bf_)   1 MPI processes</div>
                    <div>    type: fieldsplit</div>
                    <div>      FieldSplit with MULTIPLICATIVE
                      composition: total splits = 4</div>
                    <div>      Solver info for each split is in the
                      following KSP objects:</div>
                    <div>      Split number 0 Defined by IS</div>
                    <div>      KSP Object:    
                       (bf_fieldsplit_x_velocity_)       1 MPI processes</div>
                    <div>        type: preonly</div>
                    <div>        maximum iterations=10000, initial guess
                      is zero</div>
                    <div>        tolerances:  relative=1e-05,
                      absolute=1e-50, divergence=10000</div>
                    <div>        left preconditioning</div>
                    <div>        using NONE norm type for convergence
                      test</div>
                    <div>      PC Object:    
                       (bf_fieldsplit_x_velocity_)       1 MPI processes</div>
                    <div>        type: ilu</div>
                    <div>          ILU: out-of-place factorization</div>
                    <div>          0 levels of fill</div>
                    <div>          tolerance for zero pivot 2.22045e-14</div>
                    <div>          using diagonal shift on blocks to
                      prevent zero pivot [INBLOCKS]</div>
                    <div>          matrix ordering: natural</div>
                    <div>          factor fill ratio given 1, needed 1</div>
                    <div>            Factored matrix follows:</div>
                    <div>              Mat Object:               1 MPI
                      processes</div>
                    <div>                type: seqaij</div>
                    <div>                rows=16, cols=16</div>
                    <div>                package used to perform
                      factorization: petsc</div>
                    <div>                total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>                total number of mallocs used
                      during MatSetValues calls =0</div>
                    <div>                  not using I-node routines</div>
                    <div>        linear system matrix = precond matrix:</div>
                    <div>        Mat Object:      
                       (bf_fieldsplit_x_velocity_)         1 MPI
                      processes</div>
                    <div>          type: seqaij</div>
                    <div>          rows=16, cols=16</div>
                    <div>          total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>          total number of mallocs used during
                      MatSetValues calls =0</div>
                    <div>            not using I-node routines</div>
                    <div>      Split number 1 Defined by IS</div>
                    <div>      KSP Object:    
                       (bf_fieldsplit_y_velocity_)       1 MPI processes</div>
                    <div>        type: preonly</div>
                    <div>        maximum iterations=10000, initial guess
                      is zero</div>
                    <div>        tolerances:  relative=1e-05,
                      absolute=1e-50, divergence=10000</div>
                    <div>        left preconditioning</div>
                    <div>        using NONE norm type for convergence
                      test</div>
                    <div>      PC Object:    
                       (bf_fieldsplit_y_velocity_)       1 MPI processes</div>
                    <div>        type: ilu</div>
                    <div>          ILU: out-of-place factorization</div>
                    <div>          0 levels of fill</div>
                    <div>          tolerance for zero pivot 2.22045e-14</div>
                    <div>          using diagonal shift on blocks to
                      prevent zero pivot [INBLOCKS]</div>
                    <div>          matrix ordering: natural</div>
                    <div>          factor fill ratio given 1, needed 1</div>
                    <div>            Factored matrix follows:</div>
                    <div>              Mat Object:               1 MPI
                      processes</div>
                    <div>                type: seqaij</div>
                    <div>                rows=16, cols=16</div>
                    <div>                package used to perform
                      factorization: petsc</div>
                    <div>                total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>                total number of mallocs used
                      during MatSetValues calls =0</div>
                    <div>                  not using I-node routines</div>
                    <div>        linear system matrix = precond matrix:</div>
                    <div>        Mat Object:      
                       (bf_fieldsplit_y_velocity_)         1 MPI
                      processes</div>
                    <div>          type: seqaij</div>
                    <div>          rows=16, cols=16</div>
                    <div>          total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>          total number of mallocs used during
                      MatSetValues calls =0</div>
                    <div>            not using I-node routines</div>
                    <div>      Split number 2 Defined by IS</div>
                    <div>      KSP Object:      (bf_fieldsplit_Omega_)  
                          1 MPI processes</div>
                    <div>        type: preonly</div>
                    <div>        maximum iterations=10000, initial guess
                      is zero</div>
                    <div>        tolerances:  relative=1e-05,
                      absolute=1e-50, divergence=10000</div>
                    <div>        left preconditioning</div>
                    <div>        using NONE norm type for convergence
                      test</div>
                    <div>      PC Object:      (bf_fieldsplit_Omega_)  
                          1 MPI processes</div>
                    <div>        type: ilu</div>
                    <div>          ILU: out-of-place factorization</div>
                    <div>          0 levels of fill</div>
                    <div>          tolerance for zero pivot 2.22045e-14</div>
                    <div>          using diagonal shift on blocks to
                      prevent zero pivot [INBLOCKS]</div>
                    <div>          matrix ordering: natural</div>
                    <div>          factor fill ratio given 1, needed 1</div>
                    <div>            Factored matrix follows:</div>
                    <div>              Mat Object:               1 MPI
                      processes</div>
                    <div>                type: seqaij</div>
                    <div>                rows=16, cols=16</div>
                    <div>                package used to perform
                      factorization: petsc</div>
                    <div>                total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>                total number of mallocs used
                      during MatSetValues calls =0</div>
                    <div>                  not using I-node routines</div>
                    <div>        linear system matrix = precond matrix:</div>
                    <div>        Mat Object:      
                       (bf_fieldsplit_Omega_)         1 MPI processes</div>
                    <div>          type: seqaij</div>
                    <div>          rows=16, cols=16</div>
                    <div>          total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>          total number of mallocs used during
                      MatSetValues calls =0</div>
                    <div>            not using I-node routines</div>
                    <div>      Split number 3 Defined by IS</div>
                    <div>      KSP Object:    
                       (bf_fieldsplit_temperature_)       1 MPI
                      processes</div>
                    <div>        type: preonly</div>
                    <div>        maximum iterations=10000, initial guess
                      is zero</div>
                    <div>        tolerances:  relative=1e-05,
                      absolute=1e-50, divergence=10000</div>
                    <div>        left preconditioning</div>
                    <div>        using NONE norm type for convergence
                      test</div>
                    <div>      PC Object:    
                       (bf_fieldsplit_temperature_)       1 MPI
                      processes</div>
                    <div>        type: ilu</div>
                    <div>          ILU: out-of-place factorization</div>
                    <div>          0 levels of fill</div>
                    <div>          tolerance for zero pivot 2.22045e-14</div>
                    <div>          using diagonal shift on blocks to
                      prevent zero pivot [INBLOCKS]</div>
                    <div>          matrix ordering: natural</div>
                    <div>          factor fill ratio given 1, needed 1</div>
                    <div>            Factored matrix follows:</div>
                    <div>              Mat Object:               1 MPI
                      processes</div>
                    <div>                type: seqaij</div>
                    <div>                rows=16, cols=16</div>
                    <div>                package used to perform
                      factorization: petsc</div>
                    <div>                total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>                total number of mallocs used
                      during MatSetValues calls =0</div>
                    <div>                  not using I-node routines</div>
                    <div>        linear system matrix = precond matrix:</div>
                    <div>        Mat Object:      
                       (bf_fieldsplit_temperature_)         1 MPI
                      processes</div>
                    <div>          type: seqaij</div>
                    <div>          rows=16, cols=16</div>
                    <div>          total: nonzeros=64, allocated
                      nonzeros=64</div>
                    <div>          total number of mallocs used during
                      MatSetValues calls =0</div>
                    <div>            not using I-node routines</div>
                    <div>    linear system matrix = precond matrix:</div>
                    <div>    Mat Object:     1 MPI processes</div>
                    <div>      type: seqaij</div>
                    <div>      rows=64, cols=64, bs=4</div>
                    <div>      total: nonzeros=1024, allocated
                      nonzeros=1024</div>
                    <div>      total number of mallocs used during
                      MatSetValues calls =0</div>
                    <div>        using I-node routines: found 16 nodes,
                      limit used is 5</div>
                    <div>Number of SNES iterations = 2</div>
                  </div>
                  <div><br>
                  </div>
                  <div>I will try with 3.5.2.</div>
                  <div><br>
                  </div>
                  <div>  Thanks,</div>
                  <div><br>
                  </div>
                  <div>    Matt</div>
                  <div> </div>
                  <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">Thanks.<span><font
                        color="#888888"><br>
                        Anton<span class="HOEnZb"><font color="#888888"><br>
                          </font></span></font></span></blockquote>
                </div>
                <span class="HOEnZb"><font color="#888888"><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
                  </font></span></div>
            </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>
    </blockquote>
    <br>
  </body>
</html>