Work on interpolation from cell to face quadrature points #10870
Conversation
| template <bool do_evaluate, bool add_into_output> | ||
| static void | ||
| interpolate(const internal::MatrixFreeFunctions::ShapeInfo<Number> &data, | ||
| const Number * input, | ||
| Number * output, | ||
| const bool do_gradients, | ||
| const unsigned int face_no) | ||
| { | ||
| AssertDimension(n_points_1d, data.data.front().quadrature.size()); |
There was a problem hiding this comment.
Since this function is almost the same as the other FEFaceNormalEvaluationImpl, I wonder whether we can unify the calls and let both function call an internal function interpolate_generic. What I'm thinking about is to let interpolate and interpolate_quadrature extract the data array (shape_data_on_face[face_no % 2] or quadrature_data_on_face[face_no % 2]) as well as the in_stride and out_stride and let the interpolate_generic (or whatever name we choose) do the job.
There was a problem hiding this comment.
Good idea. What I was thinking about as a follow-up PR is to make the function dimension independent since I want to use it in 6D. What do you think?
| * Collects all data of 1D nodal shape values (that defined by the | ||
| * support of the quadrature rule) evaluated at the point 0 and 1 |
There was a problem hiding this comment.
| * Collects all data of 1D nodal shape values (that defined by the | |
| * support of the quadrature rule) evaluated at the point 0 and 1 | |
| * Collects all data of 1D nodal shape values (defined by the Lagrange | |
| * polynomials in the points of the quadrature rule) evaluated at the point 0 and 1 |
| quadrature_data_on_face[0].resize(quad.size() * 3); | ||
| quadrature_data_on_face[1].resize(quad.size() * 3); | ||
|
|
||
| dealii::FE_DGQArbitraryNodes<1> fe_quad(quad); | ||
|
|
||
| for (unsigned int i = 0; i < quad.size(); ++i) | ||
| { | ||
| Point<1> q_point; | ||
| q_point[0] = 0; | ||
| quadrature_data_on_face[0][i] = fe_quad.shape_value(i, q_point); | ||
| q_point[0] = 1; | ||
| quadrature_data_on_face[1][i] = fe_quad.shape_value(i, q_point); | ||
| } |
There was a problem hiding this comment.
I think you need to encapsulate this part in some if for the case the number of quadrature points is very high, say a few hundreds, look at what we do elsewhere in this function for the auxiliary FE. In that case, the FE_DGQArbitraryNodes polynomials would underflow and result in a division by zero. The collocation is not really helpful in that case anyway.
peterrum
force-pushed
the
quadrature_data_on_face
branch
2 times, most recently
from
c55b518
to
e94bc78
Compare
peterrum
force-pushed
the
quadrature_data_on_face
branch
from
e94bc78
to
113a314
Compare
|
Done :) |
Extracted from #10830. In the ECL context, values are available at cell quadrature points. The face quadrature points can be simply computed via a 1D interpolation as proposed here.