[SciPy-user] Numerical minimization

James R. Phillips scipy at zunzun.com
Wed Jul 21 04:57:50 CDT 2004

> The damping parameters should be positive
> d_i>0 for all i. How can I enforce this constraint ?

The function to be minimized could return a very
large value if any parameter is outside some set
of constraints, these can be anything you wish
including minimum or maximum parameter values.

> How do I find a global minimum ?

You will need to start the minimizers with an
initial set of function parameters that are,
from the minimizer's point of view, relatively
close to the actual global minimum.  This would
normally be done in one of two ways:

1) Supply good guesses as to the initial values

2) Have the software find good guesses to same

I myself use a genetic algorithm, specifically
Differential Evolution, to search for initial
parameter estimates if they are not user-supplied.

     James Phillips

More information about the SciPy-user mailing list