ufl team mailing list archive

[Martin Sandve Alnæs <martinal@xxxxxxxxx>]

2009/3/23 Anders Logg <logg@xxxxxxxxx>:
> On Mon, Mar 23, 2009 at 10:53:20AM +0900, Garth N. Wells wrote:
>>
>>
>> UFL wrote:
>> > One or more new changesets pushed to the primary ufl repository.
>> > A short summary of the last three changesets is included below.
>> >
>> > changeset:   752:f7cd7676b697d8662d2aae8e9cf0cf68ce0dfee3
>> > tag:         tip
>> > user:        "Martin Sandve Alnæs <martinal@xxxxxxxxx>"
>> > date:        Fri Mar 20 14:18:43 2009 +0100
>> > files:       ufl/algorithms/analysis.py
>> > description:
>> > Changed estimate_quadrature_order to
>> > return 2*v.degree() for linear forms.
>> >
>>
>> For nonlinear problems it's important that the quadrature order of the
>> linear and bilinear forms is the same.
>>
>> Garth
>
> It seems to me that estimate_quadrature_order() should be removed from
> UFL since quadrature has nothing to do with the form language.

It's a utility, intended to implement a standardized
way of estimating the necessary integration order.

> Instead, UFL can have a function called polynomial_order() that
> returns the polynomial order of an integrand if possible (if it is
> indeed a polynomial) and otherwise returns None (if the integrand
> contains something like exp or sin).

This is the aptly named estimate_maximum_quadrature_order
(or something like that) with the FIXME in it :)
polynomial_order is better.

> Then the form compilers can call polynomial_order() and then decide
> what to do with the result.

That was the idea anyway, estimate_quadrature_order
should just provide one standardized way to pick an order,
while polynomial_order provides a different approach.
Then the form compiler decides what to use.
I see no need to implement different strategies for this.

Martin