Thanks -
A = {10,30,50,70,90,110,130,150,170,190,210,230,250,27 0,290,310,330,350} ;
angles in degrees.
f(A) =
{0.87,0.8,1,1,0.98,0.98,0.98,0.75,0.65,0.69,0.97,0 .98,1,0.98,0.95,0.73,0.77,0.87} ; observations.
model f(A) = a0 + a1.cos A + ... + an.cos n.A + ...
ie Cos matrix . F(a) matrix = f(A) matrix
F(a) = {a0, a1, ... an, ...}
find F(a)


"Dana DeLouis" wrote:


Hi. Using LU Decomposition can sometimes be a little more stable.
If also offers a few other benefits.

http://en.wikipedia.org/wiki/LU_decomposition

Feel free to send me your data
= = =
Dana DeLouis



David C wrote:
Thank you all - what I found was that Minverse was giving silly results, or
even ERROR messages, which could be changed by small changes in some of the
'A' matrix numbers; ie the fault may well lie in the source data not the
process. I wanted an alternative (and more compact) process to compare
results, to determine where the fault may lie.
Using Minverse to solve the Cosine matrix for a cyclic model, sometimes the
resulting coefficients were divergent too, adding to the instability. The
results, when reconstructed and compared with the data, did not even pass
through the data points.
f(A) = a0 + a1.cos A + ... + an.cos n.A + ...
F(a) = [ a0, a1, ... an, ... ]
Cos matrix . F(a) matrix = f(A) matrix

Still working the problem - grateful for comments.

"Dana DeLouis" wrote:

Could be just a high "Condition Number" that's associated with his matrix.

http://en.wikipedia.org/wiki/Matrix_condition_number

Dana DeLouis


Gary''s Student wrote:
No guarantee at all!

Clearly David is having problems with some of his matrices. The code in
wikipedia is small enough that it could be re-coded in a couple of hours.

David could compare the results from the Gaussian method to whatever
Microsoft uses.