dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

On 12/08/11 02:52, Johan Hake wrote:
> Any objections to me merging this into trunk?
> 
> 
> Will add unittest for all methods soonish.
> 

How soon?

> 
> The code now resides in:
> 
> 
> lp:~dolfin-core/dolfin/hake
> 
> 
> Also, what would the best way to make it work in parallel. The distance
> from all vertices in a mesh to the closest boundary might not be easy to
> compute in Parallel as some vertices residing in one mesh might have the
> closest distance in the mesh on another processor.
> 

With difficulty. Sounds like an operation that is inherently not parallel.

> 
> I am inclined to think that this is a bad side effect a user need to be
> aware of when using this function in parallel. But then I know someone
> who would disagree ;)
> 

Add a line to throw an error in parallel until we figure out what's best.

> 
> Another thing is that the present implementation takes a GenericVector
> representing the return values of the distances at each vertex. Somthing
> like:
> 
> 
> distances = Function(V)
> 
> 
> # Compute distance to Boundary for each vertex
> 
> distance_computation = DistanceToBoundaryComputation(mesh)
> 
> distance_computation.vertex_distances(distances.vector())
> 
> 
> In vertex_distances() I check that the local size of the passed vector
> has the same size as the mesh.num_vertices() this gives an error when
> running in parallel:
> 
> 
> Expected a vector with the same local size as the number of vertices
> (1449) in the mesh, got 1427
> 
> 
> Expected a vector with the same local size as the number of vertices
> (1457) in the mesh, got 1441
>


I would suggest not using Vector. It's really for linear algebra
operations. Use std::vector or dolfin::Array.

> 
> I suspect that it has something to do with shared vertices. How do I
> access the "correct" number of vertices of a mesh partition and how do I
> know which one is only owned by local mesh?
>


Boundary vertices are not 'owned' by any one processes. Dofs on a
partition boundary are owned by one process. This information is
available via the DofMap.

> 
> I figure I have to look into ParallelData, which btw is not wrapped to
> Python. We need to add it to dolfin_mesh.h. Will do later...
> 

Don't rush to expose it. It shouldn't be required in user code.

Garth

> 
> Johan
> 
> 
> On Wednesday August 10 2011 14:02:34 Johan Hake wrote:
> 
>> Hello!
> 
>>
> 
>> I have created a class, DistanceToBoundaryComputation, similar to
> 
>> IntersectionOperator, which takes a Mesh or a FacetFunction and compute
> 
>> distances between any point and the closest point to the Boundary or a
> 
>> subset of the boundary given by the FacetFunction.
> 
>>
> 
>> I have published it together with two demos, cpp and python to illustrate
> 
>> some of its functions.
> 
>>
> 
>> lp:~johan-hake/dolfin/distance-to-boundary
> 
>>
> 
>> If the distances from each vertex is computed, will the result be similar
> 
>> (not always equal) to the signed distance function, or the eikonal
> 
>> equation, but it computes faster.
> 
>>
> 
>> I am not sure how this best can be integrated into the present dolfin. It
> 
>> generates some data, like a BoundaryMesh, which it need to store, so it
> 
>> might not be something we want to put into the Mesh class? If I use the
> 
>> same lazy initialization that Andre used it might be possible.
> 
>>
> 
>> Please feel free to have alook at one of the demos to see it in action.
> 
>>
> 
>> Johan
> 
>>
> 
>> _______________________________________________
> 
>> Mailing list: https://launchpad.net/~dolfin
> 
>> Post to : dolfin@xxxxxxxxxxxxxxxxxxx
> 
>> Unsubscribe : https://launchpad.net/~dolfin
> 
>> More help : https://help.launchpad.net/ListHelp
> 
> 
> 
> 
> _______________________________________________
> Mailing list: https://launchpad.net/~dolfin
> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~dolfin
> More help   : https://help.launchpad.net/ListHelp