Do Kakutani's, Helly's, and Goldstine's theorems all hold for general normed linear spaces?

210 Views Asked by At

I'm reading the proof of Kakutani's theorem in Brezis's book of Functional Analysis.

Theorem 3.17 (Kakutani). Let $E$ be a Banach space. Then $E$ is reflexive if and only if $$B_{E}=\{x \in E ;\|x\| \leq 1\}$$ is compact in the weak topology $\sigma\left(E, E^{\star}\right)$.

The direction "$\rightarrow$" relies on Banach–Alaoglu theorem, while the reverse "$\leftarrow$" relies on Helly's and Goldstine's theorems. All three theorems (Kakutani's, Helly's, and Goldstine's) are stated for Banach spaces. However, I could not see how completeness is used in the proofs. I think that they hold for general normed linear spaces.

I posted the full proofs from the textbook below. Could you confirm if my understanding is correct?


Proof of Kakutani's theorem. Assume first that $E$ is reflexive, so that $J\left(B_{E}\right)=B_{E^{\star \star}}$. We already know (by Banach-Alaoglu theorem) that $B_{E^{\star \star}}$ is compact in the topology $\sigma\left(E^{\star \star}, E^{\star}\right)$. Therefore, it suffices to check that $J^{-1}$ is continuous from $E^{\star \star}$ equipped with $\sigma\left(E^{\star \star}, E^{\star}\right)$ with values in $E$ equipped with $\sigma\left(E, E^{\star}\right)$. In view of Proposition 3.2, we have only to prove that for every fixed $f \in E^{\star}$ the map $\xi \mapsto\left\langle f, J^{-1} \xi\right\rangle$ is continuous on $E^{\star \star}$ equipped with $\sigma\left(E^{\star \star}, E^{\star}\right)$. But $\left\langle f, J^{-1} \xi\right\rangle=\langle\xi, f\rangle$, and the map $\xi \mapsto\langle\xi, f\rangle$ is indeed continuous on $E^{\star \star}$ for the topology $\sigma\left(E^{\star \star}, E^{\star}\right)$. Hence we have proved that $B_{E}$ is compact in $\sigma\left(E, E^{\star}\right)$.

The converse is more delicate and relies on the following two lemmas:

Lemma 3.3 (Helly). Let $E$ be a Banach space. Let $f_{1}, f_{2}, \ldots, f_{k}$ be given in $E^{\star}$ and let $\gamma_{1}, \gamma_{2}, \ldots, \gamma_{k}$ be given in $\mathbb{R}$. The following properties are equivalent:

(i) $\forall \varepsilon>0 \exists x_{\varepsilon} \in E$ such that $\left\|x_{\varepsilon}\right\| \leq 1$ and $$ \left|\left\langle f_{i}, x_{\varepsilon}\right\rangle-\gamma_{i}\right|<\varepsilon \quad \forall i=1,2, \ldots, k, $$ (ii) $\left|\sum_{i=1}^{k} \beta_{i} \gamma_{i}\right| \leq\left\|\sum_{i=1}^{k} \beta_{i} f_{i}\right\| \quad \forall \beta_{1}, \beta_{2}, \ldots, \beta_{k} \in \mathbb{R}$.

Proof of Helly's theorem. (i) $\Rightarrow$ (ii). Fix $\beta_{1}, \beta_{2}, \ldots, \beta_{k}$ in $\mathbb{R}$ and let $S=\sum_{i=1}^{k}\left|\beta_{i}\right|$. It follows from (i) that $$ \left|\sum_{i=1}^{k} \beta_{i}\left\langle f_{i}, x_{\varepsilon}\right\rangle-\sum_{i=1}^{k} \beta_{i} \gamma_{i}\right| \leq \varepsilon S $$ and therefore $$ \left|\sum_{i=1}^{k} \beta_{i} \gamma_{i}\right| \leq\left\|\sum_{i=1}^{k} \beta_{i} f_{i}\right\|\left\|x_{\varepsilon}\right\|+\varepsilon S \leq\left\|\sum_{i=1}^{k} \beta_{i} f_{i}\right\|+\varepsilon S . $$ Since this holds for every $\varepsilon>0$, we obtain (ii).

(ii) $\Rightarrow$ (i). Set $\gamma=\left(\gamma_{1}, \gamma_{2}, \ldots, \gamma_{k}\right) \in \mathbb{R}^{k}$ and consider the map $\varphi: E \rightarrow \mathbb{R}^{k}$ defined by $$ \varphi(x)=\left(\left\langle f_{1}, x\right\rangle, \ldots,\left\langle f_{k}, x\right\rangle\right) $$ Property (i) says precisely that $\gamma \in \overline{\varphi\left(B_{E}\right)}$. Suppose, by contradiction, that (i) fails, so that $\gamma \notin \overline{\varphi\left(B_{E}\right)}$. Hence $\{\gamma\}$ and $\overline{\varphi\left(B_{E}\right)}$ may be strictly separated in $\mathbb{R}^{k}$ by some hyperplane; i.e., there exists some $\beta=\left(\beta_{1}, \beta_{2}, \ldots, \beta_{k}\right) \in \mathbb{R}^{k}$ and some $\alpha \in \mathbb{R}$ such that $$ \beta \cdot \varphi(x)<\alpha<\beta \cdot \gamma \quad \forall x \in B_{E} . $$ It follows that $$ \left\langle\sum_{i=1}^{k} \beta_{i} f_{i}, x\right\rangle<\alpha<\sum_{i=1}^{k} \beta_{i} \gamma_{i} \quad \forall x \in B_{E}, $$

and therefore $$ \left\|\sum_{i=1}^{k} \beta_{i} f_{i}\right\| \leq \alpha<\sum_{i=1}^{k} \beta_{i} \gamma_{i} $$ which contradicts (ii).

Lemma 3.4 (Goldstine). Let $E$ be any Banach space. Then $J\left(B_{E}\right)$ is dense in $B_{E^{\star \star}}$ with respect to the topology $\sigma\left(E^{\star \star}, E^{\star}\right)$, and consequently $J(E)$ is dense in $E^{\star \star}$ in the topology $\sigma\left(E^{\star \star}, E^{\star}\right)$.

Proof of Goldstine's theorem. Let $\xi \in B_{E^{\star \star}}$ and let $V$ be a neighborhood of $\xi$ for the topology $\sigma\left(E^{\star \star}, E^{\star}\right)$. We must prove that $V \cap J\left(B_{E}\right) \neq \emptyset$. As usual, we may assume that $V$ is of the form $$ V=\left\{\eta \in E^{\star \star} ;\left|\left\langle\eta-\xi, f_{i}\right\rangle\right|<\varepsilon \quad \forall i=1,2, \ldots, k\right\} $$ for some given elements $f_{1}, f_{2}, \ldots, f_{k}$ in $E^{\star}$ and some $\varepsilon>0$. We have to find some $x \in B_{E}$ such that $J(x) \in V$, i.e., $$ \left|\left\langle f_{i}, x\right\rangle-\left\langle\xi, f_{i}\right\rangle\right|<\varepsilon \quad \forall i=1,2, \ldots, k . $$ Set $\gamma_{i}=\left\langle\xi, f_{i}\right\rangle$. In view of Lemma $3.3$ it suffices to check that $$ \left|\sum_{i=1}^{k} \beta_{i} \gamma_{i}\right| \leq\left\|\sum_{i=1}^{k} \beta_{i} f_{i}\right\|, $$ which is clear since $\sum_{i=1}^{k} \beta_{i} \gamma_{i}=\left\langle\xi, \sum_{i=1}^{k} \beta_{i} f_{i}\right\rangle$ and $\|\xi\| \leq 1$.

Proof of Theorem 3.17, concluded. The canonical injection $J: E \rightarrow E^{\star \star}$ is always continuous from $\sigma\left(E, E^{\star}\right)$ into $\sigma\left(E^{\star \star}, E^{\star}\right)$, since for every fixed $f \in E^{\star}$ the map $x \mapsto\langle J x, f\rangle=\langle f, x\rangle$ is continuous with respect to $\sigma\left(E, E^{\star}\right)$. Assuming that $B_{E}$ is compact in the topology $\sigma\left(E, E^{\star}\right)$, we deduce that $J\left(B_{E}\right)$ is compact-and thus closed一in $E^{\star \star}$ with respect to the topology $\sigma\left(E^{\star \star}, E^{\star}\right)$. On the other hand, by Lemma 3.4, $J\left(B_{E}\right)$ is dense in $B_{E^{\star \star}}$ for the same topology. It follows that $J\left(B_{E}\right)=$ $B_{E^{\star \star}}$ and thus $J(E)=E^{\star \star}$.