interior facet integrals do not work on RT element

[Jan Blechta]

from dolfin import *

mesh = UnitSquareMesh(4, 4)
RT1 = FunctionSpace(mesh, 'RT', 1)

u = Function(RT1)
v, w = TrialFunction(RT1), TestFunction(RT1)

assemble(dot(u, u)('+')*dS) # Pass
assemble(dot(u, w)('+')*dS) # Fail
assemble(dot(v, w)('+')*dS) # Fail

Both fails with

error: ‘detJ’ was not declared in this scope

[Colin Cotter]

The following (which arises from reconstructing fluxes for DG advection) also fails:

n = FacetNormal(mesh)

V1 = FunctionSpace(mesh,"RT",1)
v = TestFunction(V1)
u = TrialFunction(V1)

mat = assemble(inner(v('+'),n('+'))*inner(u('+'),n('+'))*dS)

[Marie Elisabeth Rognes]

This is a bug in tensor representation, in other words it runs with quadrature representation:

form_compiler_parameters={"representation": "quadrature"}
assemble(..., form_compiler_parameters=form_compiler_parameters)

Also, here is the pure UFL example to help with debugging

RT1 = FiniteElement('RT', triangle, 1)
u = Coefficient(RT1)
w = TestFunction(RT1)
L = dot(u, w)('+')*dS # Fail

[Marie Elisabeth Rognes]

Fix committed

[Prof Garth Wells]

When changing status to resolves, it's helpful to add the changeset id.

[Marie Elisabeth Rognes]

Good point. Fixed in this:

commit f09815da164cd3741bfd00f3ef5c36c4563b5286
Author: Marie E. Rognes <[e-mail]>
Date:   Mon Mar 10 21:28:12 2014 +0100

    Implement restrictions for monomial determinants.

[Prof Garth Wells]

Thanks - it you just put the commit ID in the message, Bitbucket is clever enough to automatically make it a link to the change.

[Martin Sandve Alnæs]

Removing milestone: 1.4 (automated comment)