Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. 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 $\lambda_r$ be the Lebesgue measure restricted to the interval $[-r,r]\subset \mathbb{R}$. Each $\lambda_r$ can be normalized to $\mu_r=\lambda_r/2r$ which is a probability. The sequence $\lambda_r$ converges for $r \rightarrow \infty$ to the Lebesgue measure on the real line, but the sequence $\mu_r$ does not converge to a countably additive measure but to a finitely additive probability measure. The limit measure $\mu$ in fact satisfies $\mu(A)=0$ for any bounded subset of $\mathbb{R}$ but $\mu(\mathbb{R})=1$.

I can obtain another finitely additive probability measure by considering the sequence $\nu_r=\dfrac{1}{e^r}\int_{-\infty}^r e^x dx$, which is a probability and at the limit satisfies again $\nu(A)=0$ for any bounded subset of $\mathbb{R}$ and $\nu(\mathbb{R})=1$.

There are infinite ways to do that. My question do the two finitely additive probability measures obtained as limits of the two different sequence $\mu_r$ and $\nu_r$ are different? In which subsets of $\mathbb{R}$ do they differ? Do you know any reference that discusses the generation of finitely additive probability measures as a limit of sequences of countably additive measures?

share|cite|improve this question
What topology do you consider on the set of measures over $\mathbb{R}$? In other words, what do you mean by measures $\lambda_{r}$ converging to the Lebesgue measure? And note that it is an uncountable family of measures, not a sequence, unless you consider $r\in\mathbb{N}$ ofcourse. – T. Eskin Apr 23 '12 at 15:40
Also $\nu_{r}$, as defined above, is a constant for each $r$. What type of measure should it represent? – T. Eskin Apr 23 '12 at 15:45
$\mu\ne\nu$ since $\mu(\mathbb R^+)=\frac12\ne1=\nu(\mathbb R^+)$. – Did Apr 23 '12 at 23:36
I mean $\lambda_r(x)=1$ if $x \in [-r,r]$ and zero otherwise. You are right is not a sequence, I should specify that $ r \in \mathbb{N}$. – vatna Apr 24 '12 at 6:55
up vote 1 down vote accepted

I object to

the sequence $\mu_r$ [converges] to a finitely additive probability measure.

because it suggests we have a limit as $r\to\infty$. In reality, the following happens: every $\mu_r$ is a point of the unit sphere in $L_1(\mathbb R)$, which is a subset of the unit sphere of $L_\infty(\mathbb R)^*$. By weak* compactness, $\mu_r$ have at least one limit point in the unit sphere of $L_\infty(\mathbb R)^*$. (We can think of any element of $L_\infty(\mathbb R)^*$ as a finitely additive measure because it can be applied to characteristic functions of measurable set.) Any such limit point could be called $\mu$ and there are many of them. For example, on the set $A=\bigcup_k [4^k, 2\cdot 4^k]$ the measures $\mu_r$ take values anywhere between $1/6$ and $1/3$. By restricting to a subsequence with $\mu_{r_j}(A)=\gamma$ for a particular $\gamma\in (1/6,1/3)$ we get a finitely additive measure $\mu$ with $\mu(A)=\gamma$. This already gives uncountably many choices for $\mu$.

And whatever choice of $\mu$ and $\nu$ are made, they will not be the same, as Did pointed out: $\mu((-\infty,b])=1/2$ and $\nu((-\infty,b])=0$ for any real $b$.

For more information, search for invariant mean or read about amenable groups.

share|cite|improve this answer

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.