libadjoint team mailing list archive

[Martin Sandve Alnæs <martinal@xxxxxxxxx>]

On 13 June 2012 13:22, Patrick Farrell <patrick.farrell06@xxxxxxxxxxxxxx> wrote:
> On 13/06/12 12:15, Martin Sandve Alnæs wrote:
>>
>>
>> If added, Evaluate(u, x) would have to be replaced by a preprocessor
>> with a coefficient containing the evaluated value prior to form
>> compilation.
>>
>> This type of preprocessing opens up many possibilities and many
>> potential implementation issues I don't see clearly right now.
>>
>
> Hi Martin,
>
> Ah, thanks for the clarification. Since adding a new measure
> (as per Marie's idea) is equivalent, would it have the same
> problems?

No, that is actually a quite different feature. It will require an
addition to UFC, which splits the evaluation in a point into a global
dolfin part (finding the cell) and a local part in generated code
behind UFC (evaluating in a point local to a given cell). A symbolic
operator on the other hand, must be composable within other symbolic
expressions, such that the generated code behind UFC would need to
"reach" beyond the local cell context to get global data.

> Anyway, what I want to do isn't possible now, so I will
> change what I want to do. I think integrating over a bump
> function centred at that point should do nicely for the physical
> question I want to answer, anyway.
>
> Thanks,
>
> Patrick

No problem!

Martin