[Dune] DUNE + explicit finite volume scheme

[Oliver Sander]

Hi Nadejda!
Thanks for your interest in Dune. I'll try to answer some of your 
questions, even
though I am not quite sure I understood them correctly.

> Our problem by development of a multi-purpose solver was related to 
> the slow access (even with iterators) to the elements of numerical 
> vector implemented by using of various C++ data structures (s. Table 2 
> of the file attached: it is not the final version of the text; I send 
> it for our discussion only).
>
> We installed DUNE (EMI, Freiburg) yesterday and –according to my tests 
> – Dune::FieldVector is the only C++ data structure providing the 
> faster access than FORTRAN. I guess / hope that by coupling to DUNE 
> data structures we would be able to develop the **fast** solver.
>
Are you sure your measuring setup is correct? I can't believe that 
FieldVector is much faster than
FORTRAN code. I also cannot really believe that it is faster than a C 
array, because that is what
it is, after all. Also, I find it a bit misleading to compare static and 
dynamic containers.

> 1) At the moment I do not need the grid itself. I have an unstructured 
> grid saved in external file. I need the DUNE grid data structure (with 
> its fast access to the grid entities) , which I can fill with my mesh 
> and specify with my properties (s. Table 1 of the file attached) . How 
> to do it? My current implementation for the grid readout is shown in 
> the file read_grid.pdf attached. How to do the same and fill the grid 
> of – probably - ALUGrid type? Imagine, we have only a 3D macrogrid (no 
> refinement) consisting of hexahedrons.
>
You probably want to use the GridFactory class.
http://www.dune-project.org/doc/doxygen/dune-grid-html/classDune_1_1GridFactory.html


> 2) In your example p.43 http://www.dune-project.org/doc/grid-howto/ 
> you set only one property to each grid entity: concentration for cells 
> , velocity for face, etc. but if it is a long list of – complex 
> (cell.content: multifluid or embedded in the cell object) – properties ?
>
The Dune grids only provide you with indices (think: integers) that you 
can use to access
containers with your data. (look at the documentation for the IndexSet 
class) It is entirely up to
you which containers you use. In particular, your container may contain 
any type of data
per face/cell/vertex

> 3) I would like to compare the efficiency of the DUNE grid container 
> with boost::intrusive::unordered_set (s. Table 3). Please recommend 
> the simplest DUNE grid type, which I can fill with only one kind of 
> entities: cells , i.e. Cuboids with defined index, unit lengths and 
> volume and perform analysis as in Table 3.
>
The simplest grid is YaspGrid, which is a uniform hexahedral grid. It 
does something entirely
different from an unordered_set, though, and I don't really see how 
comparing the two would
make sense.

Hope this helps,
Oliver

> To avoid disappointments: this e-mail is not an application for a 
> joint project, but it is an attempt to find a common language for a 
> possible collaboration (if the DUNE tests will be positive).
>
> Thank you very much in advance for your support and with best regards,
>
> Nadeschda Kirchner
>
> Dr. rer. nat. Nadejda Kirchner
>
> Abteilung Numerische Simulation
>
> Fraunhofer Ernst-Mach-Institut
>
> Nadejda.Kirchner at emi.fraunhofer.de
>
> www.emi.fhg.de <http://www.emi.fhg.de>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Dune mailing list
> Dune at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune


-- 
************************************************************************
* Oliver Sander                ** email: sander at mi.fu-berlin.de        *
* Freie Universität Berlin     ** phone: + 49 (30) 838 75348           *
* Institut für Mathematik      ** URL  : page.mi.fu-berlin.de/~sander  *
* Arnimallee 6                 ** -------------------------------------*
* 14195 Berlin, Germany        ** Member of MATHEON (www.matheon.de)   *
************************************************************************