dolfin team mailing list archive

[Anders Logg <question92245@xxxxxxxxxxxxxxxxxxxxx>]

Question #92245 on DOLFIN changed:
https://answers.launchpad.net/dolfin/+question/92245

Anders Logg proposed the following answer:
On Mon, Nov 30, 2009 at 06:37:13AM -0000, Phil Marinier wrote:
> New question #92245 on DOLFIN:
> https://answers.launchpad.net/dolfin/+question/92245
>
> I am using gradients in my calculations. So far this hasn't posed a problem. I've been able to do most of my manipulation using c++ arrays. I calculated my gradients using a form file and the resulting Function has 2, (probably 3 but I'm working in 2d) SubFunctions: Gradient[0] is the X values, Gradient[1] is the Y, (and Gradient[2] is the Z?). I extract the data by doing:
>
> Function GradX; GradX = Gradients[0]; GradX.vector().get(gradX);
>
> where gradX is the c++ array. I have to do this because Gradients[0].vector().get() doesn't seem to work. Anyways, so far it hasn't been a problem. But now I need to do:
>
> GradX.vector().set(gradX); Gradients[0] = GradX;
>
> That didn't work, so now I'm trying it like this:
>
> dolfin::Function GradAvg;
> GradAvgX = &GradAvg[0];
> GradAvgY = &GradAvg[1];
>
> GradAvgX and Y are functions defined previously. And now my problem
> becomes defining a Function with 2 SubFunctions. I think this will
> work I can figure out how to define the Function, which I have no
> idea how to do. Since the programmers refernce is off the site, I
> can't look there. So this is probably a really simple and kinda
> stupid question. Sorry about that.
>
> The second part of my question involves FunctionSapces. The only
> thing I am using GradAvg for is defining Neumann BCs. As I
> understand it, I need to define the same number of points for my
> Function as there are in the domain that was used to create the
> FunctionSpace. For example, if "mesh" has 100 nodes, if I need to
> define 100 values, regardless of if the subdomain I am using this
> Function in only has 8 nodes. All the values that I don't care about
> should be set to zero, or i guess it doesn't really matter, since
> they won't be used. At least, that's the way I've been doing when
> the numbering system doesn't match up with my boundary. Because when
> nodes 0 through 7 will be assigned the first 8 values, and if those
> nodes aren't on your boundary, they won't be used.  But since it's
> now possible to define a Function without including a FunctionSpace,
> how will dolfin figure it out? I have a subdomain for my neumann
> BC. Do I just need to define values for every point in my subdomain?
> or will the Function automaticly try to find a FunctionSpace to be
> associated with? And if it does, will I then need to define a value
> for every point in my domain, instead of just on my boundary
> subdomain? And if it automaticly finds a FunctionSpace, will there
> be problems if I use the same Function in two different classes with
> different FunctionSpaces?
>
> I appologise again if this is a really stupid question. If so, if
> you could just direct me to where i can find the information, that
> would be awesome.
>
> Thank you very much for all your help,

I'm not sure what you are trying to do here. As I understand it, you
first compute the gradient (using projection?) and then you try to
change its values using set.

Manipulating the dof values of a Function directly is possible, but
it's dangerous since you need to know exactly how the dofs are
numbered.

What are you using the gradients for and why are you using a form file
to compute the gradient if you are manipulating the values manually
anyway?

--
Anders

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