Open Olender opened 4 days ago
@pbrubeck do you have any ideas?
The CG dofs are not in lexicographic ordering anymore, but we still generate quadrature rules in lexicographic ordering. In 1D we order first the vertex dofs and then the interior ones.
There is definitely an inconsistency with the DOF and quadrature orderings, I'll assign this bug to myself.
Describe the bug We were modifying our code to work with the latest Firedrake version, and every test with SEM quads now shows greater error.
Steps to Reproduce Steps to reproduce the behavior: The error initially showed in an acoustic wave propagation. Below I have a reduced version with an analytical solution:
In order to debug, I have also done a simple
u * v * dx
assemble:Expected behavior For the first code, I expected an
errornorm
close to 5e^-8, which was achieved in older versions of Firedrake. For the second code, I have used only a degree=2 element, because it is easier. According tox_func
andy_func
, the node 0 has position (0.5, 0.5) and the assembled diagonal matrix has Mdiag[0] value 0.02777. Solving for the matrix gave me 0.4444.Mdiag
values, and values I expected to see:Environment:
Output of
firedrake-status
: