Let $(\Omega,\Sigma)$ be a measurable space (no reference measure is chosen!), and $V$ a finite-dimensional normed vector space. Note carefully that I am not choosing any topology on $\Omega$, so the $\sigma$-algebra $\Sigma$ is a priori not induced by any Borel structure whatsoever.

The total variation $|\mu|$ of a $V$-valued measure is defined as $$ |\mu|(E)=\sup\limits_{\pi}\sum\limits_{E_i\in E}|\mu(E_i)|, $$ where the supremum is taken among all possible disjoint partitions $\pi=\cup E_i$ of the measurable set $E\in\Sigma$. (the set-function $|\mu|$ is always a positive measure, see [Rudin, real and complex analysis]). For an arbitrary measure we denote $$ \|\mu\|:=|\mu|(\Omega). $$ We denote $\mathcal M(\Omega)$ the set of all measures with finite total variation $\|\mu\|<\infty$, and $\mathcal M(\Omega)$ is therefore a normed vector space with the above total variation norm.

Question: is $(\mathcal M(\Omega), \|\cdot\|)$ automatically a Banach space?

When $\Sigma$ is the Borel algebra then this is true of course, because we can identify $\mathcal M(\Omega)$ with the topological dual $C_b(\Omega;V^\ast)^\ast $ and the dual of a Complete vector space is automatically complete (and in fact $\|\mu\|=\sup\limits_\phi \int \phi(x)\cdot d\mu(x)$ with $\cdot$ denoting the finite-dimensional $V,V^\ast$ pairing). However, I've never seen the statement written in this full generality, so I'm wondering whether this is actually true or not?

  • $\begingroup$ Two statements here are in need of correction. You say that the dual of a complete vector space is always complete. If you mean that the dual of a complete normed space (aka Banach space) is complete, then this is true but the completeness condition is superfluous. If you mean a complete topological vector space (even lcs), it is false. Secondly, if $X$ is ,say, a completely regular space (in particular, locally compact or metrisable), then the dual space of $C_b(X,V^\ast)$ is indeed a space of measures on the Baire $\sigma$-algebra, but not the one you are interested in. $\endgroup$ – user131781 May 22 at 15:05
  • $\begingroup$ Thanks for pointing this out, I meant indeed a Banach space, not a general topotlogical vector space. Regarding the second part of your comment: I was not aware of these subtleties. Out of curiosity: what would be the dual space (space of measures on the Baire $\sigma$-algebra) that you mentioned for completely regular spaces? $\endgroup$ – leo monsaingeon May 22 at 19:22

Indeed $(\cal{M}(\Omega),\|\cdot\|)$ is a Banach space. For $V = \mathbb{R}$ or $V = \mathbb{C}$ you can find this result in Dunford/Schwartz (1957), Linear Operators I, ch. III.7.4, in particular p. 161. For arbitrary Banach space $V$ this also holds true, but with a sligthly different norm (see p. 160). For finite dimensional $V$ this norm is equivalent to your norm. Proof as in Lemma III.1.5 of Dunford-Schwartz.

| cite | improve this answer | |
  • $\begingroup$ Yup, that does the job. (Although I find this book is a bit outdated for my taste and hard to read, nothing beats the classic old-timers). Thank you very much @Dieter. $\endgroup$ – leo monsaingeon May 22 at 11:26

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.