[proFit-list] Fitting Functions and Return Values

[Shannon Stewman]

On Jan 20, 2009, at 11:36 AM, Andrew Cameron Beveridge wrote:

> 1.) Any ideas on how to program a dirac delta function in proFit  
> that can
> be used with the derivatives procedure?

The robust fitting procedure (which I think uses something like amoeba/ 
downhill simplex) may be able to handle this, but you may be stuck  
using Monte Carlo, which is slow but can deal with discontinuities  
like a delta function.

Alternatively, you may want to play with using a sigmoidal form to  
approximate the step function, for example, sigmoid(x;a)=1/(1+exp(-a  
x)).  At large enough values of the parameter a, the behavior is close  
to a Heaviside step function, but its form has a continuous  
derivative, which lets you use Levenberg-Marquardt and avoid other  
pitfalls of discontinuous functions/derivatives.

Best of luck,

Shannon Stewman
stew at uchicago.edu