Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

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

How to prove this? Please help me. Thank you very much.

A measurable set $E$ in a measure space $(X, \mathcal{M}, \mu)$ is said to be an atom if $\mu (E) > 0$ and no proper measurable subset of $E$ has positive $\mu$ measure.

Let $(X, \mathcal{M}, \mu)$ be a $\sigma$- finite measure space. Prove that the set of extreme points in $B_{L^1(\mu)}=\{v \in L^1(\mu): \|v\| \le 1\}$ is equal to $\{\pm \mu(F)^{-1} \chi_F: F \mbox{ is an atom of }\mu\}$.

share|cite|improve this question

Under your unusual definition of an atom, this statement is false. Consider $X=[0,1]$ and $\mathcal M$ the $\sigma$-algebra generated by $X$ and its countable subsets, and let $\mu(A)=1$ if $A$ is uncountable and $\mu(A)=0$ if $A$ is countable. This $\sigma$-finite measure space has no atoms (since you can always remove further points from any set of positive measure and still leave a set of positive measure), yet $\pm\chi_X$ are extreme points of $B_{L^1(\mu)}$ (in fact the only points of $B_{L^1(\mu)}$).

The statement is true, however, under the usual definition of an atom (see e.g. Wikipedia and PlanetMath), under which an atom is a set $E$ with positive measure that has no subsets $S$ with $\mu(E)\gt\mu(S)\gt0$, i.e. all subsets of $E$ have measure either $0$ or $\mu(E)$. In the above example, $X$ is an atom under this definition, so things work out.

To prove the statement under this definition, first let $f$ be an extreme point of $B_{L^1(\mu)}$, and let $S$ be its support. If $S$ is not an atom, we can write it as a union $S=S_1\cup S_2$ of sets of positive measure. The integral of a positive integrable function over a set of positive measure is positive. (To see this, consider the preimages of $(1,\infty)$ and $(1/(n+1),1/n]$, which cannot all have measure zero.) Thus we can write $f$ as $f=I_1f_1+I_2f_2$ with

$$I_k=\int_{S_k}|f|\,\mathrm d\mu$$


$$f_k=\frac{f\cdot\chi_{S_k}}{I_k}\in B_{L^1(\mu)}\;,$$

contradicting the fact that $f$ is an extreme point. Hence $S$ is an atom. (Note that this argument wouldn't work under your definition of an atom, since in that case one of $S_1$ and $S_2$ could have zero measure, and then one of $I_1$ and $I_2$ would be zero and this wouldn't be a proper convex combination.)

Since $f$ is measurable, it must be constant on $S$ except for a set of zero measure: The intersections of the preimages $f^{-1}([x,\infty))$ with $S$ must all have measure $0$ or $\mu(S)$, so there must be some value $x_0$ were the jump occurs, and then $\mu(f^{-1}(\{x_0\}))=\mu(S)$. Thus $f$ is equivalent in the $L^1$ sense to a constant function on $S$, and since $\lVert f\rVert=1$, the constant must be $\pm\mu(S)^{-1}$. Thus all extreme points of $B_{L^1(\mu)}$ are of the given form.

I won't spell the other direction out in detail so as to leave some of your homework for you – if $\pm\mu(F)^{-1}\chi_F$ with $F$ an atom is a convex combination, you can show that the integrals over the absolute values of the summands outside $F$ must vanish, so the support of the summands is $F$ plus perhaps a set of measure zero, and by the above argument the summands have to be constant on $F$.

share|cite|improve this answer
Can you please elaborate more in the paragraph before the end? I don`t see why f must be constant over S. – Danny Apr 23 '12 at 4:07
@Danny: There was a slight inaccuracy in that paragraph, which I've now corrected. Since $f$ is measurable, the preimages of the measurable sets $[x,\infty)$ must be measurable, and so must their intersections with $S$. Since $S$ is an atom, the measure of the intersections can only be $0$ or $\mu(S)$. This measure is a non-increasing function of $x$. For any non-increasing function of $x$ with exactly two values, there must be a point $x_0$ at which the function jumps from one value to the other. – joriki Apr 26 '12 at 11:47
If $\mu(f^{-1}([x_0,\infty))\cap S)=\mu(S)$ and $\mu(f^{-1}([x,\infty))\cap S)=0$ for all $x\gt x_0$, it follows that $\mu(f^{-1}(\{x_0\})=\mu(S)$ and $\mu(f^{-1}(\mathbb R\setminus\{x_0\})\cap S)=0$. Thus $f$ differs from a function constant on $S$ only on a set of measure zero, and it is therefore equivalent in the $L^1$ sense to a function constant on $S$. – joriki Apr 26 '12 at 11:47
Thanks a lot for your nice explanations, but I have another question. If I want to read more about extreme points and convex geometry, do you know any nice book for beginners? Any book? – Danny Apr 26 '12 at 22:09

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.