p-adaptation finalized for multi-material

[adityakpandare]

In addition to @WeizhaoLi2018's p-adaptive multimaterial changes, this change-set also removes the old least-squares (LS) reconstruction functions that used only the face-neighbors. Why:

1. It was tested that this choice of neighbor-stencil can be unstable. The better choice is to use node-neighbors, which is expensive due to the large neighborhood, but very stable.
2. It wasn't being used.
3. The upkeep of this LS scheme was turning out to be more than initially thought.

Removing the face-neighbor LS means that compflow and transport cannot work with scheme="p0p1" anymore.

---

This change is 

[WeizhaoLi2018]

Reviewed 75 of 75 files at r1, all commit messages.
Reviewable status: all files reviewed, 5 unresolved discussions (waiting on @adityakpandare)

---

src/PDE/DGPDE.hpp line 199 at r1 (raw file):

                           tk::Fields& prim,
                           std::size_t nielem,
													 std::vector< std::size_t >& ndofel ) const

Tab format issue

---

src/PDE/PrefIndicator.cpp line 91 at r1 (raw file):

  auto epsH = std::pow(10, -4 - tolref * 4.0);
  auto epsL = std::pow(10, -6 - tolref * 8.0);
  //auto epsL_p2 = std::pow(10, -7 - tolref * 8.0);

Here the epsL_p2 as well as the "else if" part below are left here since I observed that sometimes just use epsL for both DGP2 and DGP1 might not produce good ndof distributions. You can perform some further testings by manupulating these three thresholds.

---

src/PDE/PrefIndicator.cpp line 400 at r1 (raw file):

  }

  Ind = dU / U;

Do I remove this indicator based on velocity?

---

src/PDE/Reconstruction.cpp line 35 at r1 (raw file):

void
lhsLeastSq_P0P1( const inciter::FaceData& fd,

Do we remove all these functions or transfer into other places?

---

tests/regression/inciter/transport/GaussHump/CMakeLists.txt line 19 at r1 (raw file):

                    LABELS alecg)

add_regression_test(gauss_hump_p0p1 ${INCITER_EXECUTABLE}

Do we remove all these regression testing?

[adityakpandare]

Reviewable status: all files reviewed, 5 unresolved discussions (waiting on @WeizhaoLi2018)

---

src/PDE/DGPDE.hpp line 199 at r1 (raw file):

Previously, WeizhaoLi2018 (Weizhao Li) wrote…

Tab format issue

Done.

---

src/PDE/PrefIndicator.cpp line 91 at r1 (raw file):

Previously, WeizhaoLi2018 (Weizhao Li) wrote…

Here the epsL_p2 as well as the "else if" part below are left here since I observed that sometimes just use epsL for both DGP2 and DGP1 might not produce good ndof distributions. You can perform some further testings by manupulating these three thresholds.

Thanks for the note. I made a comment in the code to this effect.

---

src/PDE/PrefIndicator.cpp line 400 at r1 (raw file):

Previously, WeizhaoLi2018 (Weizhao Li) wrote…

Do I remove this indicator based on velocity?

Yes, the mm_padap branch removed velocity-based indicator. I was also curious about that. Do you remember why?

---

src/PDE/Reconstruction.cpp line 35 at r1 (raw file):

Previously, WeizhaoLi2018 (Weizhao Li) wrote…

Do we remove all these functions or transfer into other places?

Removed all of these. They are never used for multimat. Only for compflow, which can also use the extended stencil version below.

---

tests/regression/inciter/transport/GaussHump/CMakeLists.txt line 19 at r1 (raw file):

Previously, WeizhaoLi2018 (Weizhao Li) wrote…

Do we remove all these regression testing?

Yes, because we removed the face-neighbor p0p1 least-squares reconstruction, I removed the testing for it. To put it back, we will have to update the baselines with the node-neighbor least-squares. I can try to do that eventually when I get a chance, but for now, this will have to go.

[WeizhaoLi2018]

Reviewed all commit messages.
Reviewable status: 73 of 75 files reviewed, 1 unresolved discussion (waiting on @adityakpandare)

---

src/PDE/PrefIndicator.cpp line 400 at r1 (raw file):

Previously, adityakpandare (Aditya Pandare) wrote…

Yes, the mm_padap branch removed velocity-based indicator. I was also curious about that. Do you remember why?

If I removed it, it must be based on the results that showing this does not contribute to the final Ind we are using.