3

I have some questions about Brezis book. We know that M is reflexive, so there exists an homeomorphism $J:(M,\|\|_M) \to (M'',\|\|_{L(M',R)})$ between the "strong" topology of M and M''. (Moreover J is isometric).

  1. So i would like to know why $B_M=B_{M''}=\{\xi\in M'':\|\xi\|_{L(E',R)}\leq 1\}$ and why $(M,\sigma(M,M')=(M,\sigma(M'',M')$
  2. when the ball $B_M$ is metrizable in the weak topology $(M,\sigma(M,M')$ by consequence $B_{M''}$ is metrizable for the weak-star topology $(M,\sigma(M'',M')$ ?

enter image description here

Translation :

Theorem III.27 : Let $E$ be a reflexive Banach space, $(x_n)$ a bounded sequence in $E$. Then there exists a sub-sequence extracted from $(x_n)$ which converges for the $\sigma(E,E')$ topology

Demonstration : Let $M_0$ be the vector space generated by the $(x_n)$, et $M=\overline{M_0}$. M is a separable space (see III.23). Furthermore, M is reflexive (see III.17). Therefore $B_M$ is a compact, metrizable space for the topology $\sigma(M,M')$. Indeed, $M'$ is separable (see III.24) hence $B_{M''}(=B_M)$ is metrizable for $\sigma(M'',M')(=\sigma(M,M'))$ (see III.25). We can then extract a sub-sequence $(x_{n_k})$ which converges for $\sigma(M,M')$. We conclude that $(x_n)$ converges too for $\sigma(E,E')$ (by restricting $M$ to linear forms on, $E$).

curious
  • 295
  • It's not just any old isometric isomorphism, it's the canonical embedding of $M$ into $M''$ that is an isometric embedding. The balls are the same only when identifying the two spaces via the canonical isometry (they lie in different spaces otherwise), but under this identification it is clear that we have equality, since $J$ is an isometry. Without the identification, it reads $J(B_M) = B_{M''}$. The same for the topologies induced by $M'$. They are only equal when identifying $M$ and $M''$ via $J$, otherwise $J$ is a homeomorphism $(M;\sigma(M,M'))\to (M'',\sigma(M'',M'))$. – Daniel Fischer Oct 05 '14 at 09:01

1 Answers1

2

We don't actually have equality, but we have a canonical identification between the two spaces. This canonical identification is habitually left implicit for notational simplicity (at the cost of temporarily confusing beginners).

With the identification made explicit, the assertions are

  1. $J(B_M) = B_{M''}$, which immediately follows from the fact that $J$ is an isometric isomorphism, and

  2. $J\colon (M,\sigma(M,M')) \to (M'',\sigma(M'',M'))$ is a topological isomorphism. In particular, the restriction of $J$ to $B_M$ is a homeomorphism between $B_M$ and $B_{M''}$, where both are endowed with the subspace topology induced by $\sigma(M,M')$ and $\sigma(M'',M')$ respectively.

It is clear that if two spaces are homeomorphic, each is compact resp. metrisable if and only if the other is.

To see that $J\colon (M,\sigma(M,M'))\to (M'',\sigma(M'',M'))$ is a topological isomorphism, consider the standard neighbourhood bases of $0$ in these topologies: Given $\mu_1,\dotsc,\mu_k\in M'$, we have

$$\begin{aligned} J\left(\{ x \in M : \lvert \mu_\kappa(x)\rvert < 1 \text{ for } 1 \leqslant \kappa\leqslant k\}\right) &= J\left(\{x \in M : \lvert J(x)(\mu_\kappa)\rvert < 1 \text{ for } 1 \leqslant \kappa\leqslant k\}\right)\\ &= \{ J(x)\in M'' : \lvert J(x)(\mu_\kappa)\rvert < 1 \text{ for } 1 \leqslant \kappa\leqslant k\}\\ &= \{\varphi\in M'' : \lvert \varphi(\mu_\kappa)\rvert < 1 \text{ for } 1 \leqslant \kappa\leqslant k\}, \end{aligned}$$

so $J$ induces a bijection between the two neighbourhood bases, and that implies that $J$ is a homeomorphism, since $J$ is linear.

Daniel Fischer
  • 206,697
  • Dear Daniel Fisher, thank you very much for your really useful and exhaustive answer! I just would like to be sure i well understood: – curious Oct 05 '14 at 13:50
  • 1-using properties of bijectivity and isometry are suficient to show equallity $J(B_M)=B_{M''}$ ?

    2- Since the properties of bijectivity and linearity are not affected by the changement of topology we just have to show bicontinuity. To show the fact that any element of the base of neigbourhood of the weak topology of M is sent in an element of the base of neighbourhood of M'' for the weak topology you focused on 0 center neighbourhood (translation) and you took the reciproque image of 1 radius balls (homotétie) (instead of any $\varepsilon>0$:is it because of linearity of our objects $\mu_k$?

    – curious Oct 05 '14 at 14:08
  • Yes, bijectivity and isometry are the conditions that ensure $J(B_M) = B_{M''}$. 2. Yes, since vector space topologies are translation invariant, one needs only consider the neighbourhoods of $0$. Generally, one needs only consider "balls" with radius $1$ for any seminorm because of the homothety-invariance, although here we can scale the $\mu_\kappa$ to achieve the same effect. And yes, it's the linearity of the $\mu_\kappa$ that allows that.
  • – Daniel Fischer Oct 05 '14 at 15:05