ufl team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

2008/9/30 Anders Logg <logg@xxxxxxxxx>:
> On Mon, Sep 29, 2008 at 11:49:31PM +0200, Martin Sandve Alnæs wrote:
>> 2008/9/29 Anders Logg <logg@xxxxxxxxx>:
>> > On Mon, Sep 29, 2008 at 11:03:14PM +0200, Martin Sandve Alnæs wrote:
>> >> I was wondering how you intended to use that :)
>> >
>> > I was planning to count the number of products each basis function
>> > appeared in. That worked but then I realized that's not what we need.
>> > We need to know how many of each basis function appears in a product.
>>
>> And that no basis functions appear as arguments to nonlinear functions like exp.
>
> Yes.
>
>> You commented that is_multilinear is FFC-specific. But a valid form must
>> always be multilinear in the basis functions, but not in functions. Right?
>
> Yes.
>
>> >> Lets discuss how to best do this soon, I can show you
>> >> how I've been thinking in the other algorithms as well.
>> >
>> > It can be easily implemented using the monomials() function I just
>> > added. It returns a list of tuples, where each tuple is a product of
>> > basis functions.
>> >
>> > The final monomials() function will do some more, like propagating all
>> > linear (differential) operators to the basis functions.
>>
>> I was considering that propagation a part of the AD implementation.
>> If we write algorithms separately we'll probably duplicate several things
>> in different but non-orthogonal ways.
>>
>> Is this monomials() approach to is_multilinear possible to generalize
>> to non-polynomial forms? Otherwise it's very FFC specific.
>
> Yes, maybe. Perhaps I should move that code to FFC?

It's difficult to know what we'll need both places. It's not a problem to have
algorithms in UFL that we don't use in both SFC and FFC though, as long
as the algorithms are general and don't have external dependencies.

> I can try to implement is_multilinear (in the sense that all forms
> must be multilinear, not just FFC forms) based on monomials(), remove
> monomials(), and put an extended version of monomials() in FFC if it's
> not something that is generally useful to have in UFL.
>
> --
> Anders

Add as much as you like to UFL, but please be careful about modifying
existing algorithms without asking me first, in case there's something
I'm using.

--
Martin