Dealing with imperfect geometry

The following discussion from !22 (merged) should be addressed:

• @inducer started a discussion: (+3 comments)

  It still doesn't work on the short stubby tube mesh, because it keeps trying to refine away on condition 1 (center is closest to its own source) and never finishes. I don't know how we can relax that condition further...

  I conjecture that this is underresolved geometry. If it is, it's not surprising that the refiner can't get rid of it.

  

  Thanks to the bits of meshmode that support grudge, I think catching this before entering the refiner is not actually hard. We should just set a tolerance on how much of a jump we allow between adjacent elements. Finding that jump (using the "opposite-face-swap" connection) then is pretty easy, at least if we're OK with doing so pointwise.

  Here's a to-do list:

  • Verify that this conjecture has a chance of being true by generating a higher-res version of the stubby tube (which should be better resolved) and checking that the problem goes away.
  • Check element-to-element jumps through the opposite-face-swap connection, to a given tolerance.
  • Relax condition 1 to include the same tolerance.

  Before we embark on this: @mattwala, do you believe this is plausible? If so, this should probably be broken out into a separate issue. Gitlab has a button for that.