ffc team mailing list archive

[Kristian Oelgaard <k.b.oelgaard@xxxxxxxxxx>]

Quoting "Garth N. Wells" <gnw20@xxxxxxxxx>:

> I've run into a problem with zero coefficients. The code:
>
>      mesh = UnitSquare(32, 32)
>
>      V = FunctionSpace(mesh, "CG", 1)
>      v = TestFunction(V)
>
>      # Source term
>      f = 0.0
>
>      # Linear form
>      L = v*f*dx
>
>      # Assemble and apply boundary conditions
>      b = assemble(L)
>
>
> gives an error since f = 0 which means that UFL eliminates it. I can get
> around it by doing
>
>    f = Function(V, "0.0")
>
> but that's not very satisfactory. It would work if I had some other term
> in the linear form which is non-zero. UFL should generate a
> representation for f = 0 if there are no other terms in the form.

Currently in UFL whenever a zero is encountered in a product, the product is
replaced by Zero. It is therefore not straightforward to 'reconstruct' the term
which resulted in the Zero since you want:

L = (0*v)*dx --> (0)*dx --> (0*v)*dx

As I see it, it can be done if we don't handle the Zeros when constructing the
products but wait until creating the integrand and then strip any Zeros in Sums
etc. (this will be less efficient) such that:

L = g*v*dx + 0*v*dx -> g*v*dx
L = 0*v*dx + 0*g*v*dx -> 0*dx

also note that we probably want:

L = 0*v*very_long_ufl_expression*dx -> 0*v*dx

for run-time efficiency.

Kristian

> Garth
> _______________________________________________
> UFL-dev mailing list
> UFL-dev@xxxxxxxxxx
> http://fenics.org/mailman/listinfo/ufl-dev
>