feasible step length in constrained maximum likelihood produces problems with gamma function


Dear all,

I'm using Constrained Maximum Likelihood (CML) and want to optimize a function which involves that the parameter is plugged into a gamma function. I have constraints in place that force the coefficients to be larger than 0 (using cml_bounds). Now, as soon as I get the message "no feasible step length found" either GAUSS crashes (this is what is happening with version12) or I get a message saying that gamma function got a non-postive parameter supplied and the program terminates (when trying it out with an earlier version). Could someone please explain what exactly happens when no feasible step is found and maybe also how to tackle the problem.



Tags: asked August 26, 2013

1 Answer


First, try setting the parameter constraint in _cml_bounds to a larger value, that is, set it further away from the constraint boundary.  Next, try other descent methods, and other line search methods as well.