This Wikipedia article gives two equivalent definitions of locally convex space (l.c.s). I don't see clearly the equivalence and I'd like to make it crystal clear.

  • Definition 1 Let $(V,\tau)$ be a TVS. It is called a l.c.s. if the origin has a local base of convex balanced absorbing sets.
  • Definitions 2 Let $(V,\tau)$ be a TVS. It is called a l.c.s. if $\tau$ is generated by a family of seminorms on $V$.

Suppose $(V,\tau)$ satisfies Definition 1. Let $\mathcal{B}$ be a local base at $0$ such that for every $C\in\mathcal{B}$, $C$ is convex, balanced and absorbing. A known result shows that the Minkowski functional $\mu_C$ is a seminorm on $V$ for $C\in\mathcal{B}$.

Question 1: Why can we conclude that $\tau$ is generated by $\{\mu_C\}_{C\in\mathcal{B}}$ so that Definition 1 implies Definition 2?

Suppose $(V,\tau)$ satisfies Definition 2 and $\tau$ is generated by a family of seminorms $\{p_\alpha\}_{\alpha\in A}$. For every finite subset $F\subset A$ and $r>0$, define $$ S_{F,r}=\bigcap_{\alpha\in F}\{x\in V:p_\alpha(x)<r\}. $$

Question 2: Why $\{S_{F,r}\}$ form a base of convex balanced absorbing sets at the origin?

Let $\tau'$ be the topology generated by $\{\mu_C\}$ in Question 1. I don't know at all why $\tau=\tau'$. For Question 2, I can show that the sets $S_{F,r}$ form a convex and balanced base at $0$. Why is each $S_{F,r}$ absorbing and are the two definitions of absorbing sets in this question the same here?


Let's begin with the last point. For a balanced set, the two definitions of "absorbing" coincide. If $A$ is balanced, and $x \in tA$ for some $t \neq 0$, then $x \in sA$ for all $s$ with $\lvert s\rvert \geqslant \lvert t\rvert$. Because $\lvert s\rvert \geqslant \lvert t\rvert > 0$ means $\bigl\lvert \frac{t}{s}\bigr\rvert \leqslant 1$, and hence $s^{-1}x = \frac{t}{s} t^{-1} x \in A$ follows from $t^{-1}x \in A$, which is the same as $x \in tA$.

Coming to question 1, we note that the $C \in \mathcal{B}$ are all neighbourhoods of $0$, so the seminorms $\mu_C$ are continuous (with respect to $\tau$). Hence the topology induced by $\{ \mu_C : C \in \mathcal{B}\}$ is coarser than $\tau$, i.e. $\tau' \subset \tau$. But, since $\mathcal{B}$ is a neighbourhood basis of $0$ (for $\tau$), given any $\tau$-neighbourhood $U$ of $0$, there is a $C\in \mathcal{B}$ with $C \subset U$. Since $C$ is by definition a $\tau'$-neighbourhood of $0$, it follows that $U$ is a $\tau'$-neighbourhood of $0$. Since vector space topologies are determined by the neighbourhood filter of $0$ it follows that $\tau'$ is finer than $\tau$. Being both coarser and finer than $\tau$ means that $\tau' = \tau$, so the topology is indeed induced by the family $\{\mu_C : C \in \mathcal{B}\}$ of seminorms.

Regarding question 2, note that each $S_{F,r}$ is a neighbourhood of $0$. And every neighbourhood of $0$ is absorbing by the continuity of scalar multiplication. For every $x\in V$, the map $s_x \colon t \mapsto tx$ is continuous, and we have $s_x(0) = 0$, so by continuity there is a $\delta > 0$ such that $\lvert t\rvert < \delta \implies tx \in S_{F,r}$. But that means $x \in u\cdot S_{F,r}$ for all $u$ with $\lvert u\rvert > \delta^{-1}$.

  • $\begingroup$ Why $C$ being heighbourhood of $0$ implies that $\mu_C$ is continuous? $\endgroup$ – Jack Dec 18 '15 at 23:04
  • $\begingroup$ Why is "$C$ by definition a $\tau'$-neighbourhood of $0$"? (I can only see that $C$ is by definition a $\tau$-neighbourhood of $0$. ) $\endgroup$ – Jack Dec 18 '15 at 23:07
  • 1
    $\begingroup$ A seminorm is continuous if and only if its unit ball is a neighbourhood of $0$. The one direction is clear, if $p$ is a continuous seminorm, then $p^{-1}([0,1)) = p^{-1}((-\infty,1))$ is open and contains $0$, hence is a neighbourhood of $0$. But that set is the (open) unit ball of $p$. Conversely, if the unit ball is a neighbourhood of $0$, then $p$ is continuous at $0$, since $p^{-1}([0,r)) = r\cdot p^{-1}([0,1))$, and the family of neighbourhoods of $0$ is stable under scaling. And the triangle inequality gives $\lvert p(x) - p(y)\rvert \leqslant p(x-y)$, which shows the continuity at $x$. $\endgroup$ – Daniel Fischer Dec 18 '15 at 23:09
  • 1
    $\begingroup$ Note that $x \in O \iff (\exists 0 < t < 1)(x \in tC)$. But since $C$ is balanced, we have $tC\subset C$ for $0 \leqslant t \leqslant 1$. That shows $O \subset C$. On the other hand, if $x\in C$, then clearly $\inf \{\lambda > 0 : x \in \lambda C\} \leqslant 1$, since $1$ is an element of the set whose infimum we take. Thus $C \subset R$. $\endgroup$ – Daniel Fischer Dec 18 '15 at 23:32
  • 1
    $\begingroup$ We start with the condition that $C$ is a neighbourhood of $0$. Since $C\subset R$, it follows that $R$ is a neighbourhood of $0$. Then $\frac{1}{2}R$ is also a neighbourhood of $0$. But $\frac{1}{2}R = \bigl\{ x\in V : \mu_C(x) \leqslant \frac{1}{2}\bigr\} \subset O$, hence $O$ is a neighbourhood of $0$. $\endgroup$ – Daniel Fischer Dec 18 '15 at 23:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.