MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $X$ be a Banach space, $X' = \mathcal{L}(X, \mathbb{K})$ its dual space. Denote by $\mathcal{B}(X)$ the $\sigma$-algebra of Borel sets and denote by $\sigma(X')$ the $\sigma$-algebra which is generated by all sets of the form $u^{-1}(C)$ for $u \in X'$ and $C \in \mathcal{B}(\mathbb{K})$.

For $X$ separable we have that

$\mathcal{B}(X) = \sigma(X')$ (*)

see e.g. "Gaussian measures in Banach spaces" by Hui-Hsiung Kuo, p. 74 - 75.

Now the author of this book does not bother to discuss the case of $X$ non-separable.

In [1] is a halfway believable counterexample for $X = \ell^2(\mathbb{R})$.

I'm specifically interested in the case $X = \ell^{\infty}$. Does (*) hold in this case and why or why not?



share|cite|improve this question
up vote 8 down vote accepted

If $I$ is uncountable, then in space $l^2(I)$ no countable set of functionals separates points. Consequently, for any set $A$ in the sigma-algebra generated by these functionals [the Baire sets for the weak topology, see reference below], if $0 \in A$, then an entire subspace is contained in $A$. So all elements of this sigma-algebra are unbounded. Thus this sigma-algebra is not all of the norm-Borel sets.

My papers on measurability in Banach space:
Indiana Univ. Math. J. 26 (1977) 663--677
Indiana Univ. Math. J. 28 (1979) 559--579


For gaussian measures in Banach space, you really want the example of Fremlin and Talagrand, "A Gaussian measure on $l^{\infty}$". Ann. Probab. 8 (1980), no. 6, 1192--1193. This gaussian measure on $l^\infty$ with the cylindrical sigma-algebra has total mass 1, yet every ball of radius 1 has measure 0.

share|cite|improve this answer
Thanks for the answer. I also downloaded your papers. That's very fascinating research. If you could give me the reference for $\ell^{\infty}$ that would be great. – santker heboln May 13 '10 at 13:29
Gerald - Does the example of a Gaussian measure on $\ell^\infty$ prove the original question either way? My guess is that the cylindrical sigma algebra on $\ell^\infty$ is not the same thing as the Borel sigma algebra (otherwise, why even use the cylindrical sigma algebra in the example). But I'm not sure that the example implies this. – George Lowther May 16 '10 at 16:12
I think it does since a Radon measure has a unique restriction to $\sigma(X')$. But I now have found a reference where it is shown directly that $\sigma(\ell^{\infty}') \not= \mathcal{B}(\ell^{\infty})$. See "Probability Distributions on Banach Spaces" by Vakhania et al, p. 23 - 24. – santker heboln May 16 '10 at 20:16
Thanks Santker, that's a nice reference which answers your original question. I'm not sure what the significance of your comment about Radon measures is though (the example wasn't Radon). – George Lowther May 17 '10 at 0:45

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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