Error with Transfinite Interpolation

[Juan Carlos Araujo Cabarcas]

Dear all,

Since the inclusion of Transfinite interpolation, I have been successful on working with this powerful technique in my research. I had coded a mesh implementing concentric circles, where the inner most is shifted a small distance s. All concentric circles are labeled 100+i, where i is in a loop on all circles.

The mesh was working after 4/Apr/17 when I added the following entry in the forum:
https://groups.google.com/forum/#!topic/dealii/hCZqv9g6mKk

I installed the development version of dealii on the 17/nov/17. The details of the installation are also in the forum:
https://groups.google.com/forum/#!topic/dealii/ee2w2X987ig

and recently I went back and tried to run thhis code, obtaining the error at the end of the email.

I prepared a minimal example that includes the definition of my grid. It includes how I colorize each face as explained before, and the rest are colorized "1" as the documentation suggests.

The symptoms are the following:
- If I use refinement=0, everything works and the mesh looks quite nice in zeros.vtk and surface with lines mode in Paraview.
- For any higher refinement, it gives the error below.

Maybe there has been a major change in TransfiniteInterpolation since Apr/2017, or maybe the way I colorize is not good anymore for some weird reason. Any ideas or comments would be greatly appreciated.

Juan Carlos Araújo,
Umeå Universitet.



terminate called after throwing an instance of 'dealii::Mapping<2, 2>::ExcTransformationFailed'
  what(): 
--------------------------------------------------------
An error occurred in line <1554> of file </path/dealii/source/grid/manifold_lib.cc> in function
    typename dealii::Triangulation<dim, spacedim>::cell_iterator dealii::TransfiniteInterpolationManifold<dim, spacedim>::compute_chart_points(const dealii::ArrayView<const dealii::Point<spacedim> >&, dealii::ArrayView<dealii::Point<dim> >) const [with int dim = 2; int spacedim = 2; typename dealii::Triangulation<dim, spacedim>::cell_iterator = dealii::TriaIterator<dealii::CellAccessor<2, 2> >]
The violated condition was:
    false
Additional information:
   
--------------------------------------------------------

Aborted (core dumped)

[David Wells]

Hey there,

I have some familiarity with the transfinite interpolation implementation and am looking into this now. It looks like, in the transformation, one of the 'reference cell points' is calculated as {20, 20}: I don't yet see why. I suspect that we now create a twisted cell (or something else that is poorly conditioned) or that there is an issue with how we compute chart points: either way, its either something we need to catch with a better error message or a bug we need to fix.

Over the last few months Martin has added a lot of optimizations to the TFI code, in addition to using it in a more general setting during mesh refinement: put another way, our implementation has changed substantially, so this problem is at least in an area of the library where we have recently made huge changes.

Thanks,
David Wells

[David Wells]

I figured it out: we have an optimization that assumes in one particular place that, if we have eight points, they must be the vertices of a cube. However, in 2D, we use eight points to place new points on quadrilaterals: assuming these are the vertices of a cube is wrong. The original assumption created a bad initial value for a quasi-newton solver, which then failed to converge.

I got rid of the cube assumption. see https://github.com/dealii/dealii/pull/5761 for more information.

Do you mind if I use a modified version of your code for a new test in deal.II?

Thanks,
David Wells

On Wednesday, January 17, 2018 at 3:05:48 PM UTC-5, Juan Carlos Araujo Cabarcas wrote:

[Juan Carlos Araujo Cabarcas]

Thanks a lot for looking at this issue so quickly!

No problem, you can use it. That was the idea of sharing the minimal example! I am always happy to contribute with the library.

So, does this mean that the way I colorize is correct? if there is a better/simpler way plz let me know.
Is the solution merged with the latest master branch? I would like to have a working library for this particular example.

[David Wells]

I have to admit that I didn't read your code carefully enough to really understand how you did the coloring. Regardless, its probably fine: using a 2D approximation for a 2D calculation gets rid of the error in the example you provided :)

--
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
---
You received this message because you are subscribed to a topic in the Google Groups "deal.II User Group" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/dealii/7rhZad4KWOw/unsubscribe.
To unsubscribe from this group and all its topics, send an email to dealii+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[David Wells]

We merged the fix this morning into the development branch, including a test based on your code.

A suggestion: on some versions of GCC, calling 'abs' results in calling the C function 'int abs(int)': to be safe, use std::abs.

Let me know if things work :)

Thanks,
David Wells