For a dual pair $\langle X, Y \rangle$ of vector spaces $X$ and $Y$ denote by $\sigma(X,Y)$ the weak topology and by $\beta(X,Y)$ the strong topology on $X$. Consider the following known statement

(S) if $X$ is normed and $\sigma(X,X') = \beta(X,X')$ then $X$ is finite-dimensional.

There are infinite-dimensional locally convex spaces $X$ with $\sigma(X,X') = \beta(X,X')$. In fact, if $Y$ is any vector space and $Y^*$ the algebraic dual of $Y$ then $\sigma(Y^*, Y) = \beta(Y^*, Y)$. As an example, if $Y = \varphi$ (the space of sequences that are eventually $0$) with its largest locally convex topology then $Y' = Y^* = \omega$ (the space of all sequences) and thus $\sigma(\omega, \varphi) = \beta(\omega, \varphi)$. In other words, $X := \omega$ equipped with the topology $\sigma(\omega, \varphi)$ is an infinite-dimensional locally convex space for which its strong topology coincides with its weak topology. But note also that $X$ is a Fréchet space.

To which types of locally convex spaces $X$ can statement (S) be extended?


For a (Hausdorff) locally convex space (l.c.s.) you have $\sigma(X,X')=\beta(X,X')$ if and only if all $\sigma(X',X)$-bounded subsets of $X'$ are finite dimensional (i.e., their linear span has finite dimension).

In general, this is a surprisingly rich class of l.c.s.: By an old result of Susanne Dierolf every l.c.s. is a quotient of such a space (which is also complete).

On the other hand, if $X$ is barrelled with $\sigma(X,X')=\beta(X,X')$ then the $\sigma(X',X)$-bounded sets are equicontinuous and this implies that that all continuous semi-norms of $X$ have co-finite dimensional kernel (so that you are very close to a product of lines).

  • $\begingroup$ Thank you. Do you possible know of some more explicit classification of dual pairs $\langle X, Y \rangle$ satisfying $\sigma(X, Y) = \beta(X,Y)$, e.g. involving direct sums or products of real lines (or complex planes)? Or is it just hopeless to write down such an explicit list? $\endgroup$
    – yada
    Apr 11 '17 at 12:46

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.