Working with fully distributed triangulation: example dealii-pft/step-5

[Aditya Pratap Singh]

Hi,

I was trying to understand dealii-pft/step-5 and implement something similar in my code, but there are some things which are confusing for me!

As from line 55, Author copies the data from serial triangulation to a fully distributed triangulation and then reinitialize it:  

// create instance of pft
parallel::fullydistributed::Triangulation<dim> tria_pft(comm);
// extract relevant information form serial triangulation
auto construction_data =
parallel::fullydistributed::Utilities::copy_from_triangulation(basetria, tria_pft);
// actually create triangulation
tria_pft.reinit(construction_data);

As I tried to implement similar, firstly I cannot find any function called reinit for triangulation file. Also, the way construction_data has been created, I was unable to find function called copy_from_triangulation for doing the same! I am assuming that this is because I have not included proper header file.

I can see that author has included <deal.II/distributed/tria_util.h> in the code, but it was not available atleast in dealii version 9.2.0 or 9.3.0 . 

Can anyone suggest something for this?

Looking forward to some valuable suggestions.

Thanks and Regards,

Aditya

[peterrum]

Hi Aditya,

what you found is the old development (and proof of concept) code. Since that time, we have integrated p:f:T into deal.II. All the tests there have been converted to proper deal.II tests in the folder https://github.com/dealii/dealii/tree/master/tests/fullydistributed_grids. I suggest you to take a look at the test copy_serial_tria_*.cc and copy_distributed_tria_*.cc. With increasing number of indices the tests become more complex (adding PBC, MG, ...). So start with 01 ;)

Hope that helps!

PM

[Aditya Pratap Singh]

Hello,

Many thanks for the reply. 

I went through these examples, it is quite helpful to understand the concept of copying a serial or distributed triangulation to fully distributed triangulation. I wanted some clarity in one of the examples (copy_serial_tria_04) of triangulation with periodic faces:

// create serial triangulation
Triangulation<dim> basetria;
GridGenerator::subdivided_hyper_cube(basetria, n_subdivisions);
// new: add periodicy on serial mesh
add_periodicy(basetria);
basetria.refine_global(n_refinements);
GridTools::partition_triangulation_zorder(
Utilities::MPI::n_mpi_processes(comm), basetria);

// create instance of pft
parallel::fullydistributed::Triangulation<dim> tria_pft(comm);
// extract relevant information form serial triangulation
auto construction_data =

TriangulationDescription::Utilities::create_description_from_triangulation(
basetria, comm);
// actually create triangulation
tria_pft.create_triangulation(construction_data);
// new: add periodicy on fullydistributed mesh (!!!)
add_periodicy(tria_pft);

As highlighted in red color, may I know why periodicity is added twice to the triangulation (before and after copying )?  And will same applies for a p:d:t basetria as well because no such example was discussed for this case?

Thanks and Regards,

Aditya

[peterrum]

The reason why you need to add periodicity twice (independent of the type of the base tria) is that during copying the mesh the periodicity is not applied. When I implemented p:f:T I did not find an easy way to do it. 

Hope this helps,

PM

[Aditya Pratap Singh]

A follow-up question will be: what if we only apply the periodicity after copying the serial mesh to p:f:t? Why it is necessary to apply periodicity before copying the mesh as well?

[peterrum]

This is needed to guarantee that the levels of neighboring cells differ at most by 1 also the periodic boundaries. This is guaranteed by default within the domain but not at the boundaries (only if pbcs have been applied).

PM