using downhill simplex routine from 'numerical recipes'

From: jj (jungmin113_at_yahoo.com)
Date: 03/29/05 

Date: Tue, 29 Mar 2005 13:57:40 EST

dear group,

i am using the downhill simplex routine from the numerical recipes for my two-variables optimization (minimization) problem. let me say that i must stick to this particular routine.

i roughly know the solution range for both variables; one variable, say A(>0), is about 0.5~1.5 and the other, say B, is about -0.05 ~ 0.05, note that |B| could be 0.0d-15, 0.001 or 0.000001 once it converged.

my problem is that depending my initial guesses of A and B i have different outcomes. i thought that the order difference causes the problem, so i normalized both variables to 1. now the problem is that variable A can be converged to negative and very small (-1.0d-5) (before normalization i didn't have this problem.) any comments?

according to the book, i need to have a characteristic length and a unit vector for extra points (vertices.)

                  [ Pi = Po + lambda * ei ]

knowing my solution range, what are my lambda and unit vectors for variables A & B? specially for variable B.

Many thanks to all of you for your time in advance....

s. yoon