Let $\mathcal{K}$ be a family of finitely additive scalar-valued measures defined on some $\sigma$-field $\Sigma$.

We say that $\mathcal{K}$ is uniformly countably additive provided for each decreasing sequence $(E_n)$ of members of $\Sigma$ with $\cap_n E_n = \emptyset$ and each $\epsilon >0$ there is an $N_{\epsilon} >0$ such that $| \mu (E_n) | \leq \epsilon$ for $n$ beyond $N_{\epsilon}$ and all all $\mu \in \mathcal{K}$.

On the other hand, assume that $\mathcal{K}$ is NOT uniformly countably additive.

The textbook says then we can choose a sequence $(\mu_n)$ in $\mathcal{K}$, a disjoint sequence $(E_n)$ in $\Sigma$ and $\delta >0$ such that $|\mu_n (E_n)| > \delta$ for every $n$.

But, I couldn't choose such disjoint sequence $(E_n)$ in $\Sigma$. Only one I could choose is a decreasing sequence whose intersection is the empty set.

Could you help me to choose such disjoint sequence?


Let $F_n$ be a decreasing sequence in $\Sigma$ (that is, $F_{n+1}\subset F_n$) with empty intersection such that $\lim(\mu(F_n))_n$ does not converge uniformly to $0$ for $\mu\in K.$

(I). Suppose $\lim_{n\to \infty}\mu(F_n)=0$ for each $\mu \in K.$

Let $r>0$ be such that $S=\{n\in \Bbb N: \exists \mu\in K\;(\mu(F_n)>r)\}$ is infinite.

Take $n(1)\in S$ and $\mu_1\in K$ with $\mu_1(F_{n(1)})>r.$

Recursively for $j\in \Bbb N$ let $n'(j) >n(j)$ such that $n\geq n'(j)\implies \mu_j(F_n)<r/2.$ Let $E_j=F(n(j))\setminus F(n'(j)).$

Recursively let $ n'(j)<n(j+1)\in S$ and let $\mu_{j+1}\in K$ such that $\mu_{j+1}(F_{n(j+1)})>r.$

If $j<k$ then $E_j\cap E_k=\phi$. And $\mu_j(E_j)>r/2$ for all $j$.

(II). (I am presently too sleepy to consider the case $\neg (\forall \mu\in K\;(\lim_{n\to \infty}\mu(F_n)=0)$. )

  • $\begingroup$ Thanks! I used your argument to prove the claim. On the other hand, we can choose a decreasing sequence $(F_n)$ with empty interior and $\epsilon >0$ such that $S = \{ n \in \mathbb{N} | \exists \mu \in \mathcal{K} : | \mu(F_n)|> \epsilon\}$ is infinite. Also, since $F_n$ is decreasing with empty intersection, $\lim_n \mu (F_n) =0$ for $\mu \in \mathcal{K}$ (sorry to forget to mention that we may assume that all $\mu \in \mathcal{K}$ are countably additive); hence we can choose such $n'(j) > n(j)$. Then by using your argument, we can find such disjoint sequence and positive $\epsilon$. Thanks! $\endgroup$ – cdamle Oct 30 '17 at 6:06
  • $\begingroup$ I'm uncertain how general the definition of "finitely additive scalar-valued measure" is here. Some def'ns would include the case $ \Sigma=\{\phi,S\}$ for some $S\ne \phi,$ with $\mu (S)=\infty$. $\endgroup$ – DanielWainfleet Oct 30 '17 at 10:32

I think you loss some property on the sequence of $\mu_k$... Since you can choose $\mu_k$ to be delta function central at the same point and then they satisfied they are not uniformly countable additive but we can not choose a sequence ($\mu_n$) in $K$ such that there is a disjoint sequence ($E_n$) in $\Sigma$ and $\delta>0$ such that $|\mu_n(E_n)|>\delta$ for every $n$.


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.