Davoche
02-02-2005, 03:28 AM
I'm tryin to use the Conjugate gradient Method for a multidimensional minimization. I use Frprmn, mnbrak, dbrent, dlinmin, f1dim and df1dim.
With a simple Quadratic form : F(x,y)= (x+2)^2+(y+2)^2+3, I find a Strong dependancy of the result to the Initial Guess.
For example, If a start with x=2 and y=2, it works well ( min. at x=-2, y=-2 and F=3 ) but with with x=2 and y=1, I find x=-2, y=-1 and F=3.8....:eek:
I have noticed that It works generally well with a symetric Initial Guess x=4 and y=4 ; x=5 and y=5 ; x=10 and y=10 ....
I think there is a HUUUUUGE Problem...
With a simple Quadratic form : F(x,y)= (x+2)^2+(y+2)^2+3, I find a Strong dependancy of the result to the Initial Guess.
For example, If a start with x=2 and y=2, it works well ( min. at x=-2, y=-2 and F=3 ) but with with x=2 and y=1, I find x=-2, y=-1 and F=3.8....:eek:
I have noticed that It works generally well with a symetric Initial Guess x=4 and y=4 ; x=5 and y=5 ; x=10 and y=10 ....
I think there is a HUUUUUGE Problem...