Let $(X,\Sigma,\mu)$ be a finite measure space (i.e., $\mu(X) < \infty$). Let $\mathcal{F}$ be the set of $\mu$-measurable functions $f:X \to \mathbb{R}$ that are bounded in $[0,1]$, so that $0 \leq f(x) \leq 1$ for all $x \in X$ and $f \in \mathcal{F}$. Is the set $\mathcal{F}$ compact with respect to the topology induced by the $L_1$ metric $d(f,g) = \int_X|f(x)-g(x)|d\mu(x)$?

  • 3
    $\begingroup$ Probably not research level, but I decided to answer anyway. Jet-lag may serve as an excuse for that --- also for any mistakes in my answer. $\endgroup$ – Andreas Blass May 20 '12 at 20:39
  • $\begingroup$ @Vladimir: a Banach space is locally compact iff it is finite dimensional, and $L^1(X,\Sigma,\mu)$ is finite dimensional iff $X$ is the union of finitely many atoms. $\endgroup$ – Pietro Majer May 20 '12 at 21:12
  • 2
    $\begingroup$ It is true that for any uniformly bounded sequence $f_n$, there exists a sequence $g_n\in{\rm conv}(f_n,f_{n+1},\ldots)$ which converges in $L^1$. (conv=set of convex combinations). This is often a good enough replacement for compactness. It has been used many times in the papers of Delbaen and Schachermayer on no arbitrage conditions in stochastic calculus (and by the same authors to prove the Bichteler-Dellacherie theorem and the Doob-Meyer decomposition). $\endgroup$ – George Lowther May 20 '12 at 22:00

Well, if $X$ is a finite set, then yes. But in the cases you probably had in mind, no. Suppose, for example, that $X$ is $[0,1]$ with Lebesgue measure, and let $f_n(x)$ be the $n$-th digit of the binary expansion of $x$. No subsequence converges, since the $L_1$ distance between any two distinct $f_n$'s is $1/2$.

| cite | improve this answer | |

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.