Re: [Libmesh-users] question about error estimation

[Manav B. <manav@u.washington.edu>]

Thanks for your response, Roy.
I would certainly be willing to contribute an example, but will not  
be able to get to it for atleast a month and a half.
I routinely use the library for analysis on mesh with 1-D, 2-D and 3- 
D elements in a 3-D space for conduction heat transfer and structural  
analysis problems. My typical problems are based on wing-box mesh. I  
had to make some changes in the Mesh class of the library to get this  
to work, but that was over a year ago, and I will have to  dig into  
the code to remind myself about it. (That should not be difficult,  
since I could just diff it against the libmesh repository).

I would not mind implementing more tailored algorithms for such mesh,  
if need be. However, at this point, I would like to look more deeply  
into the algorithms. I am wondering if the patch method would work if  
I treat the intersection as a boundary. Are there some more general  
methods in the literature for such cases? I am very new to adaptive  
analysis, hence these inquiries. I would appreciate any kind of insight.

Regards
Manav


On Mar 30, 2006, at 12:39 PM, Roy Stogner wrote:

> On Thu, 30 Mar 2006, Manav Bhatia wrote:
>
>> I am using libmesh for structural analysis with a mesh that has  
>> QUAD4 elements in intersecting planes. For example, two plates  
>> joined together to form a T-joint.
>
> How well is that working, and would you be able and willing to
> contribute a short example code?
>
> I don't think we have any code in the examples which uses d-1
> dimensional elements in a d dimensional space.  If it's something we
> support, it's a capability we should probably be doing regression
> testing for; if it's something we support imperfectly we ought to see
> how easy completing that support would be.
>
>> My code only the analysis capabilities, and I am now attempting to  
>> start using the error estimation and refinement capabilities in  
>> the library. I am curious to know whether any specific error  
>> estimation algorithm would work better for the elements at the  
>> joint, or if I can use any one of the algorithms given in the  
>> library.
>
> I wouldn't be surprised if the library shattered into pieces when
> trying to do adaptive refinement on a non-manifold mesh, frankly.  The
> assumption that each element has either 0 or 1 neighbors on each side
> is pretty deeply ingrained into the code, and at a T joint you'll have
> three elements all meeting along the same side.
>
> Off the top of my head I can't think of a good way to fix that.
> You could switch to HEX8 elements, create a new "mortar" element to
> place along the joint, or break your problem into multiple manifold
> meshes and add user constraints at the junctions... but none of that
> would be very elegant or very easy.
> ---
> Roy Stogner