1
$\begingroup$

I am not a specialist in measure theory, so excuse me if this is simple.

Let $\mu$ be a finite measure on a set $X$ (for example, the Lebesgue measure on $[0,1]$). Integrable functions on $X$ can be defined as limits (in different senses, in particular, in some definitions as uniform limits, but this does not play an important role in what I want to ask) of summable step functions $$ f=\lim_{n\to\infty}\sum_{i=1}^\infty\lambda_i^n\cdot\chi_{A_i^n}, $$ where $\lambda_i^n\in{\mathbb C}$, $A_i^n\subseteq X$.

Is it possible to control the set $\{\lambda_i^n\cdot\chi_{A_i^n};\ i,n\in{\mathbb N}\}$ in this equation so that it would not be "too big"?

For example,

Suppose $f\in L_1(\mu)$ and $|f(x)|\le 1$, $x\in X$. Do there exist a sequence of measurable sets $A_i\subseteq X$ such that

1) the set of indicators $\{\chi_{A_i};\ i\in{\mathbb N}\}$ is totally bounded (or, what is the same, precompact) in $L_1(\mu)$, and

2) $f$ belongs to the closed absolutely convex hull of the set $\{\chi_{A_i};\ i\in{\mathbb N}\}$ in $L_1(\mu)$?

$\endgroup$
4
  • 1
    $\begingroup$ Without going into detail: I think the problem is assumption 2). Is it allowed that $f$ belongs to the "closed" absolute convex hull? What about the usual elementary three-step definition of the integral: First define $\int f d\mu$ for simple functions (as your sums, but with finite sum), then for non-negative functions and then for arbitrary quasiintegrable functions? $\endgroup$ May 1, 2020 at 15:35
  • $\begingroup$ Dieter, yes, the closed absolutely convex hull, of course! Excuse me, I'll correct this. $\endgroup$ May 1, 2020 at 15:37
  • $\begingroup$ You ask first that the set $\{\lambda_i^n\cdot\chi_{A_i^n} ; i, n \in \mathbb N\}$ be not too big, but your proposed 'bigness' conditions seem to ignore the coefficients $\lambda$. Is that intentional? $\endgroup$
    – LSpice
    May 1, 2020 at 23:13
  • $\begingroup$ @LSpice I had in mind that if the condition $|f(x)|\le 1$ is not fulfilled, then $\chi_{A_i}$ must be multiplied by some $C>0$. $\endgroup$ May 2, 2020 at 4:10

1 Answer 1

2
$\begingroup$

Attempt number 2. Consider the case $f\ge 0$.

For $\alpha\in[0,1]$ let $A_{\alpha}=\{x\in X, f(x)\ge \alpha\}$, which is measurable. For $n\in \mathbb{N}$ define $f_n=\frac{1}{n}\sum_{k=1}^{n}\chi_{A_{\frac{k}{n}}}$. It is easy to see that $f_n\le f\le f_n+ \frac{1}{n}$, from where convex combinations of $\chi_{A_{\frac{k}{n}}}$ converge to $f$. We only have to show that $\mathcal{A}=\{\chi_{A_\alpha}\}$ is totally bounded.

Let $\alpha, \beta\in [0,1]$. If $\alpha\ge\beta$ then $A_{\alpha}\subset A_{\beta}$, from where $\|\chi_{A_\alpha}-\chi_{A_\beta}\|_1=\mu(A_{\beta})-\mu(A_{\alpha})$. Similarly, if $\alpha\le\beta$, then $\|\chi_{A_\alpha}-\chi_{A_\beta}\|_1=\mu(A_{\alpha})-\mu(A_{\beta})$, and so in both cases $\|\chi_{A_\alpha}-\chi_{A_\beta}\|_1=|\mu(A_{\alpha})-\mu(A_{\beta})|$.

Hence, $\chi_{A_\alpha}\to\mu(A_{\alpha})$ isometrically maps $\mathcal{A}$ into a totally bounded space $[0,\mu(X)]$, and so $\mathcal{A}$ is itself totally bounded.

In the case $f$ is complex-valued, decompose $f=g-h+i(p-q)$, where $0\le g,h,p,q\le 1$, and so each of them belongs to the closed convex hulls of a totally bounded collection of indicators.

However, you cannot get general $f$ into a single convex hull. Consider $X=\{-1,1\}$ and $f$ - identity map. There are just three non-zero indicators available: $\chi_{X}, \chi_{\{1\}}, \chi_{\{-1\}}$. Assume that $f$ is in the (closed) absolute convex hull of these three elements. If $|\alpha|+|\beta|+|\gamma|\le 1$, and $f=\alpha\chi_{X}+\beta\chi_{\{1\}}+\gamma\chi_{\{-1\}}$, then $1=f(1)=\alpha+\beta$, and $-1=f(-1)=\alpha+\gamma$, and so $2=f(1)-f(-1)=\beta-\gamma\le |\beta|+|\gamma|\le 1$.

$\endgroup$
7
  • $\begingroup$ Why $\sum_{i=1}^{m_n}|\lambda_i^n|\le 1$? $\endgroup$ May 1, 2020 at 22:14
  • $\begingroup$ erz, I did not understand your reasoning about the mapping $\chi_{A_\alpha}\mapsto\mu(A_\alpha)$, but independently on this, the family $\{\chi_{A_\alpha}\}$ is totally bounded in $L_1(\mu)$ because in each sequence $\{\chi_{A_{\alpha_i}}\}$ we can choose a subsequence $\{\chi_{A_{\alpha_{i_k}}}\}$ that has limit in $L_1(\mu)$. $\endgroup$ May 2, 2020 at 7:01
  • $\begingroup$ Are you saying that this is not true when $f$ has different signs at different points? $\endgroup$ May 2, 2020 at 7:04
  • $\begingroup$ @SergeiAkbarov I've edited the answer to address both of your comments $\endgroup$
    – erz
    May 2, 2020 at 18:12
  • $\begingroup$ erz, isn't it necessary to prove that the image of the map $\chi_{A_\alpha}\mapsto \mu(A_\alpha)$ is closed, for applying what you say about the isometry? $\endgroup$ May 2, 2020 at 18:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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