How to scale up the good variable for minimization algorithm

Question

I am using quasi newton algorithm(Numerical recipes) for minimization in which i have to assign the variables. I selected some variables but i do not know those are good scale up variables. Is anybody knows how to scale up the good variable. I have a c-code with function(f(x)) and its gradients and also quasi newton algorithm code.

Thank you in advance.

Answer 1

I think in most cases you don't need scaling. But if you have the case of parameters differing by orders of magnitude, scaling maybe worthwhile. In that case I would scale the parameters by the first guess. E.g. if you can specify the starting point and it is not terribly far off of the expected true value, it should also provide reasonable scaling.