# Lagrangian (constrained) optimization - rate of change of optimal values

in constrained Lagrangian optimization what is a general way to figure out how the optimal point varies with respect to parameters of the constraints? For example maximize x*y when x + y < k, and I wanted to find dx*/dk and dy*/dk where (x*,y*) is the optimal point. How would I do that?

-
The example problem you have given has no solution since you can take x and y to be arbitrarily large negative numbers. Or do you also have a non-negativity constraint? – Jyotirmoy Bhattacharya Sep 25 '10 at 5:53

The optimal point may not be differentiable at a value of $k$ where a constraint goes from binding to non-binding. At any other value of $k$ just differentiate with respect to $k$ the (equality) first order conditions for the constraints which bind and solve for $\partial x^*/\partial k$ etc.
I would just add that even in the case where a constraint goes from binding to nonbinding that the optimal point ought to have both one-sided derivatives with respect to $k$ existing. They just won't be equal to each other (and so the derivative itself won't exist). – Mike Spivey Oct 19 '10 at 19:47