Interior point maximization on non-smooth surface

I am estimating a state-space model using maximum likelihood, and i am also doing simulations. For the actual maximisation part i have found that Interior point is the fastest way and also more accurate than BFGS, however my function to be maximized is very "bumpy" locally, overall it does behave nicely but if you soom in there are alot of local max and mins, so i figure meaby it would help if the nummerical derivatives were calculated over a greater length of area(as oposed to just moving slightly around the estimates to get them).

Is this possible ?

Thank you

[Sayyed Ahmad]

You can also devlp your algorithm in C++ inline function to accelerating your code use it by mex compiler in matlab. You will be surprised. An inline function in C++ looked like this:

template<class T> inline T plus2(T a,T b) { return a+b; }

The following Link is a very easy example to understanding mex:

http://matlab.izmiran.ru/help/techdoc/matlab_external/ch04cre5.html

have fun Ahmad

[John D'Errico]

This is not a question of speed, of compiling a mex function at all, and all you did was resurrect a 6 year old dead question.