Вы находитесь на странице: 1из 2

Proof of the Envelope Theorem for Constrained Optimisation

Corrections to Dr Ian Rudy (http://people.pwf.cam.ac.uk/iar1/contact.html) please.

This document sets out a proof of the Envelope Theorem in the case of constrained
optimisation with equality constraints. It is an expanded version of Charles Roddie's
proof, given in lectures.

Say we seek to maximise

f x1 , x 2 , , x n , [1]

subject to

g j x1 , x 2 , , x n , c j j 1 m [2]

by choosing the xi . Here, is a parameter.

Denote the optimal value of f by f * and the optimal value of xi by xi* .

Let the "value function" be defined by

v f * x1* , x 2* , , x n* , [3]

dv
We seek to find the value of . Well, differentiating [3] gives
d

dv df * x1* , x 2* , , x n* ,
[4]
d d

Using the chain rule

dv f * f * dx *
* i [5]
d i xi d

But we know that, at the optimium

L
0 for all i [6]
x i

where L is the Lagrangian L f j g j c j . So


j

f * g j
x i*
j j x * 0 for all i
[7]
i
Putting this into [5]

dv f * g j dxi*
j * [8]
d i j x i d

The order of the sums does not matter, so we can also write this as

dv f * g j dx *
j * i [9]
d j i xi d

But we also know that at the optimum, the constraint is satisfied, so

g j x1* , x 2* , , x n* , c j j 1 m [10]

And differentiating this with respect to gives:

dg j g j g j dxi*
* 0 j 1 m [11]
d i xi d

Hence from [9]

dv f * g j
j [12]
d j

So


f * j g j
dv

j
[13]
d

And hence

dv L*
[14]
d

Вам также может понравиться