Let $X$ be a set and $\beta$ be a sigma algebra of subsets of $X$. Let $\{\mu_n\}$ be a sequence of positive measure on $(X, \beta)$ such that $\mu_{n+1}(E) \ge \mu_n(E)$. Define $\mu (E):=\lim_{n \rightarrow \infty }\mu_n (E)$ prove that $\mu$ is a measure on $\beta$. Is the conclusion still true if $\mu_{n+1}(E) \le \mu_n(E)$?

I figured the bigger problem is to show that $\mu$ is countably additive. This is what I did.

Let $E_i$ be a sequence of disjoint measurable sets and $E=\cup_{i=1}^{\infty}E_i$. Want to show that $$\mu\left(\cup_{i=1}^{\infty}E_i\right)=\sum_{i=1}^{\infty}\mu\left(E_i\right).$$ Without loss of generality, assume $\mu_n(E_i)<\infty , \forall n$. then,$$\mu(\cup_{i=1}^{\infty}E_i)=\lim_{n \rightarrow \infty}\mu_n(\cup_{i=1}^{\infty}E_i)$$ since $\mu_n$ is measurable $$\mu_n(\cup_{i=1}^{\infty}E_i)=\sum_{i=1}^{\infty}\mu_n(E_i)$$. $\Rightarrow \mu(\cup_{i=1}^{\infty}E_i)=\lim_{n \rightarrow \infty}\sum_{i=1}^{\infty}\mu_n(E_i)$.

Now this is where my difficulty lie, I know I cannot just write $$\lim_{n \rightarrow \infty}\sum_{i=1}^{\infty}\mu_n(E_i)=\sum_{i=1}^{\infty}\lim_{n \rightarrow \infty}\mu_n(E_i)$$ without justification. Can Monotone convergence Theorem be used here and how? if not, can someone show me how?

  • 1
    $\begingroup$ Yes, the monotone convergence justifies it. View the series as an integral with respect to the counting measure. $\endgroup$ – Daniel Fischer Dec 3 '16 at 20:37
  • $\begingroup$ Okay, could you write it out for me. And also does it hold if $\mu_{n+1}(E)\le \mu_n(E)$ $\endgroup$ – J. Kyei Dec 4 '16 at 1:26
  • $\begingroup$ @J.Kyei In which book did you find this exercise ? $\endgroup$ – 129492 Jun 11 '19 at 15:43
  • $\begingroup$ @129492, I'm sorry I do not know exactly from what textbook it came from, it was a question from my Real analysis class. $\endgroup$ – J. Kyei Jun 13 '19 at 15:19
  • $\begingroup$ Thank you very much. It also was an exercise in my final test $\endgroup$ – 129492 Jun 14 '19 at 4:34

In the case where $(\mu_n)_{n\in \mathbb{N}}$ is an increasing sequence of measures, $\mu = \lim\limits_{n\to \infty} \mu_n$ is a measure. If the definition of measure requires that at least one set has finite measure, then we have

$$\mu(\varnothing) = \lim_{n\to\infty} \mu_n(\varnothing) = \lim_{n\to\infty} 0 = 0,$$

so $\mu$ satisfies that condition. What remains is to show the countable additivity of $\mu$. For that, we can invoke the monotone convergence theorem for Lebesgue integrals by viewing an infinite sum as an integral with respect to the counting measure - given a sequence $(E_k)_{k\in\mathbb{N}}$ of disjoint sets in $\beta$, let $f_n(k) = \mu_n(E_k)$, then if $\zeta$ is the counting measure on $\mathbb{N}$ we have

$$\sum_{k = 0}^\infty \mu_n(E_k) = \int_{\mathbb{N}} f_n\,d\zeta,$$

and the monotone convergence theorem gives

$$\mu\biggl(\bigcup_{k = 0}^\infty E_k\biggr) = \lim_{n\to\infty} \sum_{k = 0}^\infty \mu_n(E_k) = \lim_{n\to\infty} \int_{\mathbb{N}} f_n\,d\zeta = \int_{\mathbb{N}} \lim_{n\to\infty} f_n\, d\zeta = \sum_{k = 0}^\infty \mu(E_k).$$

We can also appeal to the monotone convergence theorem for series (which is a special case of the monotone convergence theorem for Lebesgue integrals, but the series version is often proved separately before integration is treated). By monotonicity, $\lim\limits_{n\to\infty} \sum\limits_{k = 0}^\infty \mu_n(E_k)$ exists (in $[0,+\infty]$), and since $\mu_n(E_k) \leqslant \mu(E_k)$ for every $k$ and $n$, we have

$$\lim_{n\to\infty} \sum_{k = 0}^\infty \mu_n(E_k) \leqslant \sum_{k = 0}^\infty \mu(E_k).\tag{$\ast$}$$

But for every fixed $K$, we have

$$\sum_{k = 0}^K \mu(E_k) = \lim_{n\to\infty} \sum_{k = 0}^K \mu_n(E_k) \leqslant \lim_{n\to\infty} \mu_n(E_k),$$

and so

$$\sum_{k = 0}^\infty \mu(E_k) = \lim_{K \to \infty} \sum_{k = 0}^K \mu(E_k) \leqslant \lim_{n\to\infty} \sum_{k = 0}^\infty \mu_n(E_k),$$

which together with $(\ast)$ shows the desired equality.

Things are different if $(\mu_n)_{n\in \mathbb{N}}$ is a decreasing sequence of measures. Then $\mu = \lim \mu_n$ need not be a measure. Consider the measures on $\mathbb{N}$ given by

$$\mu_n(A) = \sum_{k \in A} a_k^{(n)},$$


$$a_k^{(n)} = \begin{cases} \frac{1}{(k+1)^2} &, k \leqslant n \\ \frac{1}{k+1} &, n < k.\end{cases}$$

Then $\mu_n(\mathbb{N}) = +\infty$ for all $n$, but

$$\sum_{k = 0}^\infty \mu(\{k\}) = \sum_{k = 0}^\infty \frac{1}{(k+1)^2} = \frac{\pi^2}{6} < +\infty = \mu(\mathbb{N}).$$

The failure is analogous to the possible failure of continuity from above of a measure while continuity from below always holds. If $A_n \supset A_{n+1}$ for all $n$, then we need not have

$$\lambda\biggl(\bigcap A_n\biggr) = \lim_{n\to\infty} \lambda(A_n),$$

but we have this if there is an $n$ with $\lambda(A_n) < +\infty$. In the same way, for a decreasing sequence $(\mu_n)$ of measures on $(X,\beta)$, the limit $\mu$ is a measure if there is an $n$ such that $\mu_n(X) < +\infty$. (This is a sufficient, but not necessary condition.)


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.