Necessary and sufficient condition for global minima in convex optimization

79 Views Asked by At

Minimize $f(x)$ subject to $x \in C$, $C$ is convex set

Necessary and sufficient condition for $\bar x$ to be global minima of the above problem is $<\nabla f(\bar x), x- \bar x> \geq 0 $ $\forall x \in C$

proof in textbook:

$ x \in C, \bar x, \implies \bar x + \lambda (x- \bar x) \in C \forall \lambda \in (0,1) \\ f(\bar x + \lambda (x- \bar x)) \geq f(\bar x) \\ $ [what if global min is not in C? as showin below in diagram]

by taylors expansion we get

$<\nabla f(\bar x), x- \bar x> \geq 0 $ $\forall x \in C$

It also looks intuitive when i think as below

BUt what if $\bar x$ is not in C ie., when i consider C not containing global min then $\bar x + \lambda (x- \bar x) $ may not be in C right?

enter image description here