Re: [Libmesh-devel] FE::inverse_map() not converging...

[Kirk, B. \(JSC-EG\) <ben...@na...>]

Which type of elements are these?

For Quads the zero point corresponds to the centroid of the element,
which should be as good a guess as any.  For triangles, the reference
element is an isosceles triangle with the right angle at the origin,
hence the zero point is at a node.  Probably not the best place,
especially if that angle on the physical element approaches (or exceeds)
180 degrees.

-Ben

=20

-----Original Message-----
From: lib...@li...
[mailto:lib...@li...] On Behalf Of Derek
Gaston
Sent: Wednesday, June 21, 2006 12:05 PM
To: lib...@li...
Subject: [Libmesh-devel] FE::inverse_map() not converging...

Hey guys... thought I would get this out there before our meeting today
so you have a chance to think about it.

The newton iteration in FE::inverse_map() is not converging for me after
I move nodes with my smoothing stuff.  Here's what I'm getting:

############
WARNING: Newton scheme has not converged in 31 iterations:
   physical_point=3D(x,y,z)=3D(       0, 0.459733,        0)
   physical_guess=3D(x,y,z)=3D(0.115107, 0.343355,        0)
   dp=3D(x,y,z)=3D( 15.3103, -19.1814,        0)
   p=3D(x,y,z)=3D(-0.862915, -0.356698,        0)
   error=3D24.5424
WARNING:  diff is 0.233629
 point=3D(x,y,z)=3D(       0, 0.459733,        0)
 local=3D(x,y,z)=3D(0.164291, 0.293627,        0)
 lref=3D (x,y,z)=3D(-0.862915, -0.356698,        0)
##########

I get a _whole bunch_ of these.

I have tried to up the number of newton iterations (from 10 to 30) but
that didn't seem to help... so it is definitely not converging.
What's weird is that it doesn't seem to be really affecting my
solutions... to the eye they look good!

After looking through the inverse_map() code a bit it looks like we are
taking an initial guess of the zero point... it is also mentioned that a
better choice might be the centroid.  Do you think this could have some
effect on me?  Obviously my elements must not be the "good"
elements it mentions in the comments ;-)

We can talk about this more this afternoon.... I just wanted to get the
juices flowing.

Later,
Derek


_______________________________________________
Libmesh-devel mailing list
Lib...@li...
https://lists.sourceforge.net/lists/listinfo/libmesh-devel