dolfin team mailing list archive

[Anders Logg <question141904@xxxxxxxxxxxxxxxxxxxxx>]

Question #141904 on DOLFIN changed:
https://answers.launchpad.net/dolfin/+question/141904

    Status: Open => Answered

Anders Logg proposed the following answer:
On Wed, Jan 19, 2011 at 10:39:03AM -0000, Raphael Kruse wrote:
> Question #141904 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/141904
>
>     Status: Answered => Open
>
> Raphael Kruse is still having a problem:
> Hi,
>
> thank you for your answer!
>
> Yes, you are right. The final integral is over a 2-dimensional domain,
> while the trial and testfunctions are 1-dimensional.
>
> But I hoped one can calculate the desired matrix in a perhaps iterative fashion.
> Say we have q(x,y) = q(x -y) such that
>
> [Qu](x) = \int q(x - y) u(y) dy
>
> is nothing more but the convolution of q and u.
>
> Is it possible to represent [Qu](x) as function instance in Fenics for a
> given trial function u such that
>
> L = Qu*v*dx
> c = assemble(L)
>
> and c is the column of my desired matrix, which is related to my trial
> function u?

I don't see how that would work. Your L (and c) would then be
something that depends on x. We can only assemble numbers (scalars,
vectors, matrices).

Perhaps if you express your q as a sum of basis functions? Then one
could translate the dependency on x into a dependency on an index so
that the above would be assembled into a matrix? Then another trick or
two might solve the problem... I'm not trying to be cryptic, but I
don't have time to think this through in detail atm.

--
Anders

-- 
You received this question notification because you are a member of
DOLFIN Team, which is an answer contact for DOLFIN.