Re: [Libmesh-users] Blood Vessel Adaptive Mesh

[Roy S. <roy...@ic...>]

On Wed, 6 Jun 2018, Caleb M Phillips wrote:

> On Wed, Jun 6, 2018 at 1:45 PM, Roy Stogner <roy...@ic...> wrote:
>
>       How expensive depends on your time stepping, and on your topology, I
>       believe.  My biggest question would be about the latter: do you really
>       need to delete old nodes and add new nodes with every step?  If your
>       mesh topology can remain unchanged then you can simply change the
>       geometry by setting new positions for the same old nodes, which ought
>       to be much simpler.
> 
> That sounds much simpler, my only concern would be after too much
> movement if the elements would become skewed. Or perhaps some
> elements would be 100 microns and some 1 micron.

Hmm... if you're working in 3D and you're worried about a little skew
then you might be able to get away with just a smoother pass on
internal nodes.

If you're worried about a lot of skew then you'll be stuck working
with Tets.  You can't easily keep Hexes from skewing given enough
deformation, but you can't remesh them easily *either*.

> I would have to delete nodes if a cell were to die, obviously then
> it would no longer be a part of the vessel. 

This is where meeting in person will help; I'm only 90% confident I
understand what you're getting at without pictures.

We can handle element deletion (and addition) to a mesh
mid-simulation.

> Is the error always the default? I'm surprised insert_node doesn't
> take both a Node * and id.

The documentation for insert_node literally says "Primarily intended
for use with the mesh_inserter_iterator, only use if you know what
you are doing..."  I admit that that's only *slightly* scarier than
the documentation for add_node() or add_point(), but still add_*() are
intended for expert user code and insert_node() is basically only for
internal use.  If C++ had access control lists rather than just
public/protected/private then insert_node would definitely be
intra-library-only.

> Is this done through adding a point first
> with an id and then adding a node? I believe I've overthought this
> processes several times!

You can add_node(n), in which case we simply append n to the mesh
with the next unused id, or you can insert_node(n), in which case we
assume you've already chosen an unused n->id() and we scream at you if
it isn't unused, or you can add_point(p), in which case we'll put a
Node at that point.  If you add_point(p) or add_point(p, id) and id is
free, then we'll create a new Node; if you add_point(p, id) and
there's already a node with that id then we'll just move it to p.

Where this *really* gets tricky is when you want to do it on a
distributed mesh...
---
Roy