Accessing raw data in MeshWorker?

[Scott Miller]

Hi all,

I am using MeshWorker for some explicit RKDG code.  I proceed using 2 MeshWorker::integrate loops, both with Assembler::ResidualSimple<Vector<double> >.  The first integration covers all faces in my triangulation.  The second loop computes the volume contributions, including a mass matrix which I can then invert locally and update my "solution" vector.  This seems to work out very well in that I do not need any global matrices/sparsity patterns/etc.  But, I am having problems in my implementation within MeshWorker:  On each element (in my volume integration), I need access to the local component of the previously computed/assembled residual vector, e.g. something that in a non-MW code would look like "residual(local_cell_dofs[i])".  Is there currently any way to access the raw data from the IntegrationInfo or DoFInfo constructs?  I was unable to find any interface for doing so.  If nothing exists, should we add a raw data accessor function that take either an index or name as the argument?

Thanks.

-Scott

[Bärbel Janssen]

Hi Scott,

you can add your old residual to the NamedData that you pass to your
loop and then use ist within the MeshWorker. Does that help?

Best,
Bärbel

2012/10/25 Scott Miller <miller....@gmail.com>:

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see
> https://groups.google.com/d/forum/dealii?hl=en
>
>

[Scott Miller]

Hi Barbel,

I have added it to the NamedData that I give to MeshWorker::IntegrationInfoBox. But how do I access that from within when I'm on a cell? I can get the solution values, gradients, etc at the quadrature points fine, but I'm not sure how to get the raw residual vector.

-Scott

[Bärbel Janssen]

Hi Scott,

let's assume that your NamedData is called data. Then you can do
data(data.find("solution")), or you can print it like
data.print(std::cout) to see what was added to the NamedData. This way
you can access all the data you added. Does that work?

[Scott Miller]

Hi Barbel,

Let us take a step back, and I ask: How can I access the NamedData object from a MeshWorker::IntegrationInfo object?

Thanks.

-Scott

[Bärbel Janssen]

Hi Scott,

you can get it by info.values[position of data in NamedData] and the
same with gradients and second derivatives.

[Scott T. Miller]

Hi,

I thought the info.values accessed the values of the finite element solution evaluated at the quadrature points, not the "raw" data directly.

-Scott

[Guido Kanschat]

Dear Scott,

I have written a class VectorInfo analog to IntegrationInfo which does
what you want. I attached the file and am also adding it to the
subversion archive.

Best,
Guido

[Scott T. Miller]

Hi Guido,

Many thanks! I will test it out.

-Scott

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
>
>

> <vector_info.h>

[Scott Miller]

Dear Guido,

Sorry for the delay in my testing your new class. I am having some difficulty in understanding its usage. Where would I make and/or use a VectorInfoBox? I still need the IntegrationInfo and DoFInfo in my assemble function. I also see that you have an access function ::operator(unsigned int i). If I have a VectorInfo object vec_info, and I want to do "Vector<double> my_vec = vec_info(i).block(0)", what would I be passing as the value i?

Thanks again,

-Scott

On Oct 26, 2012, at 6:07 AM, Guido Kanschat wrote:

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
>
>

> <vector_info.h>

[Guido Kanschat]

Oops, maybe I should have added a little information on its use.

It is a replacement for IntegrationInfo. Or, VectorInfoBox for IntegrationInfoBox.

Do you need the functionality of both?

Best,
Guido

[Scott Miller]

Hi Guido,

Yes, I do need the functionality of both.  I have a non-linear Jacobian that I linearize and invert on the interior of each element, and the multiply the cell vector coming from the residual of the face terms.

Thanks.

-Scott