Strict convexity and best approximations

276 Views Asked by At

Let $V$ be a normed vector space. It is said to be strictly convex if its unit sphere does not contain nontrivial segments. A subset $A \subset V$ is said to have the unicity property if for any $x \in V$, there is exactly one $x' \in A$ with $|x - x'| = \inf_{y \in A}|x - y|$.

If $V$ is strictly convex then any finite dimensional linear subspace of it has the unicity property.

Does the converse hold?

1

There are 1 best solutions below

0
On BEST ANSWER

Yes, the converse is also true. Suppose the space is not strictly convex. Let $[a,b]$ be a line segment contained in the unit sphere. The function $$t\mapsto \|(1-t)a+tb\|,\qquad t\in\mathbb R\tag1$$ is convex and is equal to $1$ on $[0,1]$. Therefore, it is greater than or equal to $1$ everywhere.

The distance from $0$ to the line (1) is realized by any point of $[a,b]$. Apply translation by $-a$ to conclude that the distance from $-a$ to the line $t\mapsto t(b-a)$ is realized by multiple points. Therefore, $V$ does not have the unicity property.