<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    Hi Jason, Klaus,<br>
    <br>
    This block-by-block variation is correlated with similar variation
    in other fluid variables (density, pressure), and this persists even
    where there is a zero velocity inflow, as well as with a uniform
    grid, and with both USM and PPM (pure hydro) solvers. Modifying the
    gravity solver from Multigrid to Multipole doesn't make a difference
    either. I'm using the FLASH Gamma EOS unit. As far as I've seen,
    there is no variation in B-field across grid cells adjacent to block
    / refinement boundaries, this only happens at the inflow edge. <br>
    <br>
    I also update the magnetic field face variables (MAG_FACE_VAR and/or
    MAGI_FACE_VAR), with no effect. Div(B) still seems to be 0. <br>
    <br>
    Any other suggestions would be great!<br>
    <br>
    Thanks,<br>
    Rukmani<br>
    <br>
    <br>
    <br>
    <div class="moz-cite-prefix">On 02/18/2016 01:30 PM, Jason Galyardt
      wrote:<br>
    </div>
    <blockquote
cite="mid:CABdF9uevM2PyrdYd_1Lz1Ofx11u0mTKEjCRHxboEA9aB6_YFKA@mail.gmail.com"
      type="cite">
      <div dir="ltr">
        <div>
          <div>
            <div>
              <div>
                <div>Hi Rukmani,<br>
                  <br>
                </div>
                I used a spatially varying wind; the velocity of the
                wind varies along the boundary, but it has a
                well-defined, time-independent form. I've also seen
                problems with more realistic B-field geometries which
                (to my horror) included step functions in the domain
                interior. I had to smooth these out to avoid unphysical
                evolution in those regions. <br>
                <br>
                I've also seen some modest increase in B-field magnitude
                for the cells adjacent to a refinement boundary. I
                haven't reported the latter previously because I haven't
                had time to figure out what's going on there. You might
                try setting lrefine_min = lrefine_max to get uniform
                refinement and see whether that helps (some of our
                group's simulations do this).<br>
                <br>
              </div>
              The block by block variation does seem strange. I would
              expect this kind of variation to be correlated with
              variation in another variable. How do the other variables
              look in the problem region?<br>
              <br>
            </div>
            Another idea: could this variation be tied to the equation
            of state? If you're using one of the supported FLASH EOS
            units, you're probably fine.<br>
            <br>
          </div>
          Regards,<br>
        </div>
        Jason<br>
        <br>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Thu, Feb 18, 2016 at 11:07 AM,
          Rukmani Vijayaraghavan <span dir="ltr"><<a
              class="moz-txt-link-abbreviated"
              href="mailto:rukmani@virginia.edu"><a class="moz-txt-link-abbreviated" href="mailto:rukmani@virginia.edu">rukmani@virginia.edu</a></a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div text="#000000" bgcolor="#FFFFFF"> Hi Jason,<br>
              <br>
              Thanks! I'm using FLASH 4.2, I'll try using 4.3 to see if
              that makes a difference. I haven't tried refining on the
              magnetic variables yet.<br>
              <br>
              For the different runtime parameters -- <br>
              <br>
              1. I've tried cfl = 0.5 and 0.8, but nothing lower yet.
              I'll check to see if that works.<br>
              <br>
              2. For the Riemann Solver, I've found HLLC to be a bit
              more dissipative than HLLD, and therefore marginally
              better at smoothing out the magnetic field at the edges.
              Ditto with second order MUSCL-Hancock over third order
              PPM.<br>
              <br>
              3. All the other runtime parameters are mostly the same. I
              don't refine on the magnetic variables, but I tried higher
              overall lrefine_min (to make sure the outer edges get
              further refined) and it didn't help -- the same
              block-based discontinuity persists.<br>
              <br>
              4. I'm using a constant wind inflow for this particular
              run. One thing I checked to see was if there was a round
              off error in reading my input variables into double
              precision arrays, and this tiny "seed" instability might
              grow, but it doesn't seem to be an issue. What is strange
              is that the value (and sign) of the initial instability
              varies block-by-block. In your simulations, did you use a
              constant wind?<br>
              <br>
              Thanks,<br>
              Rukmani
              <div>
                <div class="h5"><br>
                  <br>
                  <div>On 02/18/2016 09:38 AM, Jason Galyardt wrote:<br>
                  </div>
                  <blockquote type="cite">
                    <div dir="ltr">
                      <div>
                        <div>
                          <div>
                            <div>Hi Rukmani,<br>
                              <br>
                            </div>
                            I've had some similar issues with MHD runs.
                            You didn't mention which version of FLASH
                            you're using, but I've found the latest
                            (v4.3) to be a bit more stable than v4.2 or
                            v2.5. As for runtime parameters, found the
                            following combination to be helpful:<br>
                            <br>
                            #~~~~<br>
                          </div>
                          <div># Refine on the magnetic variables:<br>
                          </div>
                          <div>refine_var_1 = "dens"<br>
                            refine_var_2    = "magp"<br>
                          </div>
                          <div># -OR-<br>
                          </div>
                          <div># refine_var_2 = "magx"<br>
                          </div>
                          <div># refine_var_3 = "magy"<br>
                          </div>
                          <div># refine_var_4 = "magz"<br>
                          </div>
                          <div># prefer higher refinement, according to
                            magp (default refine_cutoff_X = 0.8)<br>
                            refine_cutoff_2 = 0.7<br>
                            # refine_cutoff_3 = 0.7<br>
                            # refine_cutoff_4 = 0.7<br>
                            <br>
                          </div>
                          <div># Lower CFL: between 0.25 and 0.5<br>
                          </div>
                          <div>cfl = 0.5<br>
                            <br>
                          </div>
                          <div># Use second order MUSCL-Hancock
                            reconstruction scheme<br>
                          </div>
                          <div>order = 2<br>
                          </div>
                          <div><br>
                            # I've mostly used the "hybrid" slope
                            limiter, but occasionally I've found the
                            "minmod" useful in particularly difficult
                            situations <br>
                          </div>
                          <div>slopeLimiter    = "hybrid"<br>
                            <br>
                            # use flattening (dissipative) (originally
                            for PPM)<br>
                            use_flattening    = .true. <br>
                            <br>
                            # Use high order algorithm for E-field
                            construction<br>
                            E_modification  = .true.<br>
                            <br>
                            # Update magnetic energy using staggered
                            B-fields<br>
                            energyFix       = .true.<br>
                            <br>
                            # Prolongation method (injecton_prol,
                            balsara_prol) -- Using Balsara's method is
                            particularly critical, in my experience.<br>
                            prolMethod      = "BALSARA_PROL"<br>
                            <br>
                          </div>
                          <div># For the Riemann solver, I use HLLD for
                            MHD runs, and HLLC for pure hydro runs.<br>
                          </div>
                          <div>RiemannSolver    = "HLLD"<br>
                          </div>
                          <div>#~~~~<br>
                          </div>
                          <br>
                          <div>What sort of inflow conditions have you
                            implemented? Small non-linearities in the
                            inflow can grow into large unphysical
                            features over time (I've seen this happen in
                            my own simulations). So, it's worth checking
                            your boundary condition code for undesirable
                            features. In any case, I hope this helps.<br>
                          </div>
                          <br>
                          Sean: is the E_upwind option available for the
                          unsplit MHD solver in FLASH 4.3? My
                          recollection is that it caused some problems
                          in previous versions....<br>
                          <br>
                        </div>
                        Regards,<br>
                      </div>
                      Jason<br>
                      <br>
                    </div>
                    <div class="gmail_extra"><br>
                      <div class="gmail_quote">On Wed, Feb 17, 2016 at
                        9:22 PM, Rukmani Vijayaraghavan <span dir="ltr"><<a
                            class="moz-txt-link-abbreviated"
                            href="mailto:rukmani@virginia.edu"><a class="moz-txt-link-abbreviated" href="mailto:rukmani@virginia.edu">rukmani@virginia.edu</a></a>></span>
                        wrote:<br>
                        <blockquote class="gmail_quote" style="margin:0
                          0 0 .8ex;border-left:1px #ccc
                          solid;padding-left:1ex">Hi everyone,<br>
                          <br>
                          I've come across an error when updating fluid
                          variables at the inflow edge of a wind tunnel
                          simulation. I'm running a simulation of a
                          galaxy (with active dark matter particles,
                          gas, and passive particles) in a box, whose
                          fluid is initialized to be identical to the
                          incoming wind (with vx, vy, vz = 600 km/s, 0,
                          0). There is a small error (on the order of
                          1%) when updating grid cells near the inflow
                          boundary (with both USM and PPM solvers), and
                          this error is spatially correlated with  block
                          boundaries. While this error itself is
                          tolerable as far as the density and pressure
                          go, this has bad consequences for the magnetic
                          field which grows as the wind propagates
                          through the box (see attached figure,
                          xl_boundary). This figure shows slices of Bx
                          at two timesteps (annotated with block
                          boundaries and magnetic field vectors). The
                          dynamic range of Bx in this image has been
                          reduced to highlight these discontinuities. At
                          the timesteps shown in the attached image, the
                          fluctuations in Bx are ~1%, but grow with time
                          up to order unity. I've tried a variety of
                          Riemann solvers (HLLC, HLLD, Roe, Hybrid),
                          slope limiters (mc, minmod, etc.),
                          interpolation orders, prolongation methods,
                          turning on and off specific USM switches, but
                          nothing seems to solve this issue so far. Has
                          anybody else dealt with and/or successfully
                          solved this issue?<br>
                          <br>
                          Thanks,<br>
                          Rukmani<span><font color="#888888"><br>
                              <br>
                              -- <br>
                              Rukmani Vijayaraghavan<br>
                              NSF Astronomy & Astrophysics
                              Postdoctoral Fellow<br>
                              University of Virginia<br>
                              <a moz-do-not-send="true"
                                href="mailto:rukmani@virginia.edu"
                                target="_blank">rukmani@virginia.edu</a><br>
                              <br>
                            </font></span></blockquote>
                      </div>
                      <br>
                    </div>
                  </blockquote>
                  <br>
                  <pre cols="72">-- 
Rukmani Vijayaraghavan
NSF Astronomy & Astrophysics Postdoctoral Fellow
University of Virginia
<a moz-do-not-send="true" href="mailto:rukmani@virginia.edu" target="_blank">rukmani@virginia.edu</a></pre>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
        <br>
      </div>
    </blockquote>
    <br>
    <pre class="moz-signature" cols="72">-- 
Rukmani Vijayaraghavan
NSF Astronomy & Astrophysics Postdoctoral Fellow
University of Virginia
<a class="moz-txt-link-abbreviated" href="mailto:rukmani@virginia.edu">rukmani@virginia.edu</a></pre>
  </body>
</html>