Re: Re: discontinuities in stress tensor mask in thin air

["mikeshonle" <mike@xxxxxxxxxxxxx>]

Hi, Jim & Dave,
Thanks for your responses. Sorry it's taken me a while to respond-
- doing some of these tests takes all night, and then if you find a 
mistake...

Anyway, I've been doing some more experiments with varying the mesh 
spacing, and it seems to make a relatively large difference in the 
calculated force.

In one set of experiments, I varied the mesh size in the region just 
outside the inner region (which has both magnets), and depending on 
what values I used, it seems to let more of the force 
tensor "escape", which changes the force:

Inner spacing outer spacing force
.001 .005 1.208e-3
.001 .003 0.951e-3
.001 .002 0.902e-3


I also varied the inside spacing while holding the outside constant:

.0017 .005 1.125e-3
.001 .005 1.208e-3

Is this what should be expected, or could I have something setup 
incorrectly?

Thanks again,

-Mike



--- In femm@xxxxxxxxxxxxxxx, David Meeker <dmeeker@xxxx> wrote:
> James Rabchuk wrote:
> 
> > As I understand from Dave's description of the technique, the 
mesh 
> > density serves as a type of permeability in the solution of 
> > the differential (Laplaces?) equation which sets the level 
contours, 
> > the gradient of which becomes the weighting function for the 
volume 
> > integral carried out in the air over the divergence of the stress 
> > tensor. The abrupt change in permeability between regions of 
different 
> > mesh density leads to abrupt changes in the "mask". Regions of 
higher 
> > density will have a greater percentage of the level contours.
> >
> > But how about your force values? Were they more realistic, 
smoother, etc.?
> >
> > Jim
> >
> >
> The way that things are currently set up, the program solves a 
Laplace 
> equation to get the "mask." It uses the inverse of the mesh size 
> specification sort of like a "permeability," so that more highly 
meshed 
> regions tend to contribute more to the force calculation than less 
> highly meshed regions. The lines can change directions at 
boundaries 
> between mesh densities--this is analogous to the change in 
direction of 
> flux lines at the interfaces between materials of different 
> permeabilities. I don't think that this should particularly cause 
any 
> problems with the integration.
> 
> mikeshonle wrote:
> > I'm using version 3.3a3. And what's a good practical upper
> > limit on the number of mesh nodes?
> 
> The "practical" upper limit depends on how much memory you have. 
If the 
> problem is so big that you start running off of virtual memory, 
things 
> can bog down a lot--that's really the upper limit.
> 
> Dave.
> 
> -- 
> David Meeker
> dmeeker@xxxx
> http://femm.berlios.de/dmeeker