Re: [Chama-developers] LSLR probs

[Austin M. <ami...@la...>]

On Tue, 12 Aug 2003, Michael L. Hall wrote:

> John A. Turner writes:
> > so this is pretty good news for LSLR - although it's a constant D
> > problem, it's a fairly nonortho mesh with a nonlinear solution...
> 
> Actually, this is pretty much what I would expect, after some
> thought.  The LSLR method makes a local linearity assumption (first
> order) which gets plugged into the divergence discretization (plus
> one order, according to my hunch). In my experience, this gives a
> second-order method.
> 
> The problem, of course, is that the local linearity assumption it
> makes is reasonable with constant D, but wrong with variable D. The
> question is how bad the assumption will be on variable D problems.

any luck on problem 7? that would be a good one for this.


attached is kershaw stuff for ortho methods-all methods, probs 1-5, almost 
all shapes(problem 5 only has to 64, but I thought I'd get it out since it 
probably won't be done till end of today.)

we are getting convergence problems with kershaw and LSLR-won't converge 
in 20000 iterations. shape is usually 128, but have seen 64(problem 2 w/ 
64 iterations failed to converge). maybe use preconditioner?

austin