WIP: Add anisotropy to the p-adaptive driver

Anisotropy is added to the p-adaptive driver. This is done by applying the same logic in each direction one by one rather than simultaneously.

@dmoxey, this is still very much WIP but the base logic is there, I think. I still need to add the logic for the per-direction order restriction in tris and tets and then I can actually try and see if it works. Does this so far at least make sense to you?

marked as a Work In Progress

changed title from Add anisotropy in the p-adaptive driver to WIP: Add anisotropy to the p-adaptive driver

I will need to make this a user parameter at some point.

@dmoxey, I went back to the code today to try to implement the tri/tet re-orienting. I have no idea where to start though. Any tip?

I guess the steps for a very unoptimal but working version would be something like:

• determine which triangles would need to be flipped
• recreate the triangle's geometry object by flipping edges 1<->2 and update the meshgraph
• recreate the ExpList so that you rebuild all the mappings and suchlike
• copy data back again for unaffected elements and interpolate for the affected

Thanks, I'll give that a try when I find some free time.

Ok I think I've managed to code points 1 and 2. I think points 3 and 4 were already there. I'll test this when I get the chance.

added 3 commits

• 701c0af6 - Check max P for each direction separately
• a9f34eed - Make list of triangles to invert
• ac50a030 - Only for triangles!

Compare with previous version

added 1 commit

• 70a4af85 - Rotate triangles

Compare with previous version

@dmoxey, I really want to get back to this at some point. Looking back at it (and our messages), I think I had most of it coded, just hadn't had the chance to test it yet. We could probably get a paper out of this fairly quickly.

closed

reopened

@Z.Yan: it seems you've mistakenly closed this MR. I haven't worked on it in ages but am planning to get back to it eventually.

added 1773 commits

• 70a4af85...21dc3e13 - 1772 commits from branch master
• 7f46c845 - Merge branch 'master' into feature/anisotropic-p

Compare with previous version

Big update; seems to have gone smoothly. Will work on coding this up soon.

added 1 commit

• feb4cdcd - Fix now array size calls

Compare with previous version

At least the quads (and I suppose the hexes too) are working.

Hi @dmoxey: I finally came around to testing what I coded. Unfortunately, it doesn't even get to the adaptation part because it fails in the sensor calculation, more specifically here, and I get order in 'a' direction is higher than order in 'b' direction. I coded something to avoid this in the adaptation part by rotating triangles, but I had not thought about the problem of evaluating the sensor. Any thoughts? My only idea is to rotate the triangle around and test the working direction for each of the 3 rotations but I wonder if there is a cleaner way ...

added 143 commits

• feb4cdcd...f94e6cc5 - 142 commits from branch master
• 7a3ad6ea - Merge branch 'master' into feature/anisotropic-p

Compare with previous version

changed milestone to %v5.2.0

changed milestone to %v5.3.0

I think for now this is unlikely to progress much further: I will therefore close this for now, and push up the branch to the archive/nektar repository in case we have chance to develop this further in the future.

closed