ufl team mailing list archive

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

2008/6/19 Kent-Andre Mardal <kent-and@xxxxxxxxx>:
> to., 19.06.2008 kl. 15.39 +0200, skrev Martin Sandve Alnæs:
>> We could collect them more explicitly:
>>   a = u*v*dx
>>   L = f*v*ds
>>   bc = T*u == g
>>   Axb = System(a, L, bc)
>
>
> Yes, or even
>
> Axb = System ( u*v*dx == f*vdx, T*u == g)

Even better!


>> Basically, we could do something like:
>>
>>   bc = T*u_components == expression
>>
>> Here expression is any UFL-expression, which is compiled into some
>> future extension of UFC.
>>
>> u_components will typically be:
>>   u - all values (in case of non-scalar element)
>>   u[0] - a single component (vector valued element, must be fixed integer)
>>   u[0, 1] - a single component (tensor valued element, must be fixed integers)
>>   dot(u, n) - normal component of u (handled differently depending on element!)
>>
>> We _could_ define a small set of valid expressions for u_components
>> like the list above, if we are sure we can cover everything we need.
>
> But aren't these already in UFL.

The syntax for the items in the list above, yes, and we'll reuse that.
But it doesn't make sense in this context to write f.ex.
  T*exp(f)*u == g,
so the list above is a suggestion for what we define as valid in place
of u_components.


>> Also, in many cases you'll want to just fix one point to avoid translation.
>> This kind of situation makes it difficult...
>>
>
> I guess one would the mark one node and have a subdomain which is really
> only a node (?).

That should work.


I'm convinced that we can at least cover most boundary conditions
with this, but I'd like to see how it's best implemented in the
assembler before we design the UFC extension and incorporate
this into the form compilers.

--
Martin