Convex functions and subdifferential

102 Views Asked by At

I am studying real analysis, actually, I need to prepare a seminar for my professor about convex analysis and I am using the Kenneth Davidson book called ''real analysis and application'', https://carma.newcastle.edu.au/jon/Preprints/Books/Other/TimesReal.pdf.

I need to prove the Theorem 16.7.13 at the page 595, I got the proof but there is a step that is confusing:

That is, why $\hspace{0.3cm}$ $\bigcup_{j \in J(a)} \partial f_j(a) = \operatorname{conv} \{ \partial f_j(a): j \in J(a)\} ?? $

If this information helps, I Know that $\partial f_j(a)$ is a convex set, for all $j$, and clearly that the right hand is subset of left hand, but how to show the otherwise?

1

There are 1 best solutions below

0
On

The statement in the question is incorrect, however if you 'view' the sets through a linear functional $\phi$ then the result is true.

For any linear functional $\phi$ and any set $A$ we have $\sup_{x \in A} \phi(x) = \sup_{x \in \operatorname{co} A} \phi(x)$.

Take $A= \cup_k \partial f(a_k)$.

It is clear that $\sup_{x \in A} \phi(x) \le \sup_{x \in \operatorname{co} A} \phi(x)$, just from containment.

Suppose $x \in \operatorname{co} A$, then $x = \sum_k \lambda_k a_k$, with $a_k \in A$. Choose $k_0$ such that $\phi(a_{k_0}) \ge \phi(a_k)$, then $\phi(a_{k_0}) \ge \phi(x)$, and hence we have equality.