Lagrange multiplier for non-convex optimization

238 Views Asked by At

An object is free to move within the set $S=$ { $(x, y) \in R^{2} | x+y \leq \cos (y-x) $ }.

It tries to come as close as possible to the point $\mathrm{P}=(1,1) > .$ Where should it go? In order to solve the problem, sketch the set $\mathrm{S}$.

[Hint: A linear co-ordinate transform might help you to understand the situation. Although the set $\mathrm{S}$ is NOT convex, there is only one local minimum, which is also the global one we are looking for.]

Mind: This problem is not convex. Still...

  • Sketch a qualitatively correct picture of the set $\mathrm{S}$ and the point $\mathrm{P}$
  • Explain why the constraint should be binding.
  • Write the problem in standard form with Lagrange multiplier $\lambda,$ and show that there is only a single solution which is then the desired minimum.

I wrote the optimization as

min $(x-1)^2+(y-1)^2$

s.t. $x+y-cos(y-x) \leq 0$

and used KKT condition to solve this problem, but found the conditions are hard to compute because of the existence of $cos(y-x)$. (later I checked that KKT can be applied only when $(x,y)$ is from a convex set, so here is not the case.)

I could not figure out how to do the co-ordinate transformation and sketch the set $S$ either :( Could anyone shed some light on this? Thanks!

1

There are 1 best solutions below

0
On

Hint.

Making the change in coordinates

$$ u = y+x\\ v = y-x $$

the problem becomes

$$ \min \frac 12\left((u-2)^2+v^2\right)\ \ \ \text{s. t. }\ \ \ \ u-\cos v \le 0 $$

This problem has a symmetry regarding the axis $v = 0$