1

Let $f : \mathbb{R}^n \rightarrow \mathbb{R}$ (for some $n \in \mathbb{N}^\ast$) be a continuous function and $K, K'$ be two compact subspaces of $\mathbb{R}^n$ such that $K' \subset K$.

I was wondering if there exists simple sufficient conditions on $f$ to obtain that $$\arg\max_{x \in K}f(x) \cap K' \neq \emptyset$$ and sufficient conditions on $f$ to obtain that $$\arg\max_{x \in K'}f(x) \subset\arg\max_{x \in K}f(x) \cap K'$$

1 Answers1

1

Likely not very satisfying, but a simple sufficient (and in fact necessary) condition for both is that $\max_{x\in K} f(x) = \max_{x\in K'} f(x).$ Note also that the two results you are looking for are equivalent.

Suppose $\arg\max_{x\in K} f(x) \cap K'\neq \emptyset,$ then there exists $x^*\in K'$ such that $f(x^*) = \max_{x\in K}f(x).$ We can combine this with $\max_{x\in K} f(x) \geq \max_{x\in K'} f(x),$ to obtain that $$\max_{x\in K'} f(x) \geq f(x^*) = \max_{x\in K} f(x).$$ In particular, for all $x'\in \arg\max_{x\in K'} f(x),$ we have $f(x') = \max_{x\in K'} f(x) = \max_{x\in K} f(x).$ Hence also $x'\in \arg\max_{x\in K} f(x)$ (and in particular $x'\in \arg\max_{x\in K} f(x)\cap K'$).

The converse is much easier to show (thanks to OP for noticing!). Note that by compactness, $\arg\max_{x\in K'}f(x) \neq \emptyset,$ so $\arg\max_{x\in K'}f(x) \subset \arg\max_{x\in K}f(x) \cap K'$ immediately implies that the RHS is non-empty.

Dasherman
  • 4,206
  • Thank you for your answer ! I didn't notice that the two results are equivalent.

    The converse - $\arg\max_{x \in K'}f(x) \subset \arg\max_{x \in K}f(x) \cap K' \Rightarrow \arg\max_{x \in K}f(x) \cap K' \neq \emptyset$ - seems strange to me because $\arg\max_{x \in K'}f(x)$ is nonempty, so the right-hand side of the implication is true only in the case where $\arg\max_{x \in K}f(x) \cap K' \neq \emptyset$ ?!

    – deeppinkwater Jun 08 '20 at 13:49
  • I meant "left-hand side", sorry ! – deeppinkwater Jun 08 '20 at 13:59
  • It is indeed the case that the LHS is only true if the RHS is true, but this is of course exactly the point of the two being equivalent. – Dasherman Jun 08 '20 at 14:05
  • I prefered to ask, because you said: "The converse is also quite easy to show in much the same manner", but in fact it is more a question of logic (regarding $\Rightarrow$), right ? – deeppinkwater Jun 08 '20 at 14:09
  • Yes, I realize now that this is actually much easier to show than I imagined. Thanks! – Dasherman Jun 08 '20 at 14:15