2
$\begingroup$

In this SIAM Review paper the authors are giving the following necessary condition for a point being a local maximum of a convex function:

Let $F: \mathbb{R}^n \mapsto \mathbb{R}$ be convex. If $x$ is a local maximum of $F$ over the nonempty compact set $C$, then \begin{align}\tag{FOC} \langle v - x, F'(x)\rangle \leq 0 \quad \forall v \in C. \end{align}

Moreover, the authors state that if $F$ is continuously differentiable, then any $x$ satisfying the above first-order condition will be referred to as a stationary point, and otherwise a point that satisfies the first-order condition.

What I do not understand is, what changes when a point is called a stationary point? It is still not guaranteed to be a local max, right? So what really changes when a point that satisfies the first-order conditions becomes a stationary point? What additional benefits do we get? (I am assuming this will be relevant when we analyze the convergence of the Frank-Wolfe algorithm or whatever we are using to find a local maximum, but I don't understand what changes with a point being stationary.)

Improve this question
$\endgroup$
1
  • $\begingroup$ The premise here looks a little shaky. The quote says $C$ is nonempty and compact, but is it assumed to be convex? Also, the quote starts with $F$ being convex but then refers to $F'$ (presumably the gradient of $F$). Convex functions are not automatically differentiable. Is there an explicit assumption that $F'$ exists? $\endgroup$
    – prubin ♦
    Feb 13, 2022 at 16:56

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.