On Wed, Dec 21, 2011 at 5:22 PM, James Snyder <[email protected]> wrote:
> Hi -
>
> I've started experimenting with FiPy in order to get a better
> understanding of some electrostatics problems, and have put together a
> 3D model with Gmsh to try and estimate the current flow and field
> potentials for an oblong pod in a rectangular tank with conductive
> media and terminals extending into the surrounding fluid with fixed
> potentials.  I believe that the potentials are looking fairly
> reasonable, however I seem to find some interesting discontinuities
> when I try to make estimates of the current flux being emitted from
> the terminals at the ends.  At certain points (like the two run in the
> example linked) for widths of 0.39 and 0.4 the sum all the face
> gradients scaled by conductivity and dotted with the area projections
> jumps suddenly from around 3.4e-3 to 5.6e-3 while following more
> continuous variations above and below this level:
>
> https://gist.github.com/fad276821e31439e468e
>
> I expect that the current measurements will be quite sensitive to the
> cell volumes, but the discontinuity mentioned seems to persist at 0.39
> vs 0.4 even with varying mesh resolution.  Am I doing something
> foolish or is there something simple I've missed here?  Is the issue
> related to something happening in Gmsh or FiPy? I see that the cell
> count sometimes drops a bit at the transition, but if I include 0.38
> and 0.42 the discrepancy doesn't seem proportional to the drop.

Scratch this first question.  I figured out what was happening.  My
boundary selection parameters were including the edge of the outer
volume once I dropped below 0.40.  I'd still like suggestions or
responses for the remaining questions though, if possible.

>
> A few other questions:
> If I want current density to be zero at certain boundaries normal to
> that surface (insulators), what is the correct way to do that?  I see
> the new constrain approach, but I'm just setting the faceGrad value
> vector to all zeros: "potential.faceGrad.constrain([[0], [0],
> [0]],outer_edges)" but this isn't quite right.
>
> I tried using physical surface labels in 3D to label faces, and while
> the labels came through, none none of the faces are actually tagged in
> mesh.physicalFaces.  I'm not that interested in tagging values, but
> since one way to have "internal boundaries" is to carve out a section
> of a mesh, then select specific faces of that internal hole, it would
> be nice to use the physical labels.
>
> Is there an easy way to get values for a cross-sectional plane of cell
> values on an unstructured mesh like the rectangular prism used in this
> example?  I see that I can do quite a bit of manipulation in Mayavi,
> but it would be nice to have access to something like this without an
> extensive custom routine or jumping out to a viewer/postprocessor.
>
> Also, Is there any likelihood or possibility that one might be able to
> do parallel computation on unstructured meshes using Trillinos in the
> future?  I see that currently the parallel computation is only
> supported for gridded meshes.
>
> Thanks in advance for any comments or suggestions.
>
> -jsnyder

_______________________________________________
fipy mailing list
[email protected]
http://www.ctcms.nist.gov/fipy
  [NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy]

Reply via email to