Computing sensitivities in FMINCON optimization

Hello,

I am using FMINCON for a constrained optimization problem and would like to perform a sensitivity analysis to see which variables are the most important with respect to the output. After reading a few other threads, I believe I should be using the Hessian matrix to do this, but I am unfamiliar with how to go about this. I am rather new to technical computing and would appreciate any help and direction anyone can provide. Thanks!

Holly

Answer 1

[Alan Weiss]

Perhaps the first place to look is the Lagrange multiplier structure that is optionally returned:

 [x,fval,exitflag,output,lambda] = fmincon(...)

lambda contains the Lagrange multipliers; see the output argument section of the fmincon reference page. Nonzero values of the Lagrange multipliers indicate that the constraint affects the solution. Larger absolute values of the multipliers indicate relatively more important contributions, at least locally (if you change that constraint a tiny bit, the solution changes more). There is a bit more information on the structure here.

I should probably document this better. In any case, I hope this helps.

Alan Weiss

MATLAB mathematical toolbox documentation