Let $X=\{x_1,x_2,x_3,\dots\}$ be countable. Choose $p_n\geq0$ so that $\sum_n p_n=1$. For each subset $S$ of $X$, put $\mu(S)=\sum\limits_{x_n\in S}p_n$. Prove that $\mu$ is countably additive.

Let $\{S_n\}$ be a countable family of disjoint subsets of $X$, and $S=\bigcup_n S_n$. The case $S$ is finite is trivial. So we assume $S$ is infinite. Since $\sum p_n$ converges absolutely, every rearrangement of it converges to the same limit. If each $S_n$ is finite, then we can choose $0=K_0<K_1<K_2<K_3<\cdots$ and $k_1<k_2<k_3<\dots$ such that $x_{k_j} \in S_n$ if $K_{n-1}+1\leq j\leq K_n$. Then $$\mu(S)=\sum_{x_j\in S}p_j=\sum^\infty_{j=1}p_{k_j}=\sum^\infty_{n=1}\sum^{K_n}_{j=K_{n-1}}p_{k_j}=\sum^\infty_{n=1}\mu(S_n)\text{.}$$

The case that exactly one $S_n$ is infinite is similar. How to deal with the case two or more of the $S_n$ are infinite?

  • 1
    $\begingroup$ All $p_j$'s are nonnegative, so just feel free to rearrange all the sumations. $\endgroup$
    – Kervyn
    Jun 23 at 9:03
  • $\begingroup$ @Kervyn I think we need to rearrange as a double sequence. Can you state such a theorem and proof? $\endgroup$
    – user912011
    Jun 23 at 9:48
  • 1
    $\begingroup$ That's not a big problem. You can relabel the double sequence into a new sequence and show the way you relabel doesn't matter. It's actually some discrete version of Fubini theorem. $\endgroup$
    – Kervyn
    Jun 23 at 10:26

1 Answer 1


Suppose $\sum a_n$ is an absolute convergent series of complex number.

THEOREM 1 Every rearrangement of $\sum a_n$ converges to the same sum.

THEOREM 2 If $\{n_k\}$ is a sequence of positive integers such that $1=n_1<n_2<n_3<\cdots$, then $$\sum^\infty_{k=1}\sum^{n_{k+1}-1}_{n=n_k}a_n=\sum^\infty_{n=1} a_n\text{.}$$

PROOF Partial sums the former series is a subsequence of the latter.

THEOREM 3 If $\{a_n\}$ is arranged in a double sequence $\{b_{jk}\}$ as follows: $$\begin{array}{ll} b_{11}&=a_1&b_{12}&=a_{N+1}&b_{13}&=a_{2N+1}&\cdots\\ b_{21}&=a_2&b_{22}&=a_{N+2}&b_{23}&=a_{2N+2}&\cdots\\ &\vdots&&\vdots&&\vdots&\\ b_{N1}&=a_N&b_{N2}&=a_{2N}&b_{N3}&=a_{3N}&\cdots \end{array}$$ then $\sum_{k} b_{jk}$ converges for $j=1,\dots,N$, and $\sum_j\sum_kb_{jk}$ converges to $\sum_na_n$.

PROOF The convergence of $\sum_kb_{jk}$ follows from the absolute convergence of $\sum a_n$ and Cauchy criterion. For each $K$, $$ \sum^N_{j=1}\sum^K_{k=1} b_{jk}=\sum^K_{k=1}\sum^N_{j=1}b_{jk}=\sum^K_{k=1}\sum^{kN}_{j=1+(k-1)N}a_j\text{.}$$ Letting $K\rightarrow\infty$, by THEOREM 2, we obtain $\sum^N_{j=1}\sum^\infty_{k=1}b_{jk}=\sum^\infty_{k=1}a_k$.

THEOREM 4 If $\{a_n\}$ is arranged in a double sequence $\{b_{jk}\}$ as follows: $$\begin{array}{ll} T_1:&b_{11}&=a_1&b_{12}&=a_{3}&b_{13}&=a_{6}&\cdots\\ T_2:&b_{21}&=a_2&b_{22}&=a_{5}&b_{23}&=a_{9}&\cdots\\ T_3:&b_{31}&=a_4&b_{32}&=a_{8}&b_{33}&=a_{13}&\cdots\\ &&\vdots&&\vdots&&\vdots& \end{array}$$ then $\sum_{k} b_{jk}$ converges for each $j$, and $\sum_j\sum_kb_{jk}$ converges to $\sum_na_n$.

PROOF Put $a=\sum a_n$. Let $\epsilon>0$. Choose $N$ so that $\sum^\infty_{n=N}|a_n|<\epsilon$. Plainly, it is possible to choose $K$ such that $\{a_1,\dots,a_{N-1}\}\subset T_1\cup\cdots\cup T_K$. Fix $k>K$. By THEOREM 1 and 3, there exists a subsequence $\{c_m\}$ of some rearrangement of $\{a_n\}$ such that $\sum^k_{m=1}\sum^\infty_{n=1}b_{mn}=\sum c_m$. Since $\{c_m\}$ contains $a_n$ for $n<N-1$, $$|\sum c_n-a|\leq \sum^\infty_{n=N}|a_n|<\epsilon\text{.}$$ This completes the proof.

ANSWER TO THE QUESTION By theorem 1 and 4, we can arrange $\{p_n\}$ in a double sequence $\{q_{jk}\}$ so that, for each $j$, $\mu(S_j)=\sum_k q_{jk}$. It follows then that $$\sum p_n=\sum_j\sum_k q_{jk}=\sum_j \mu(S_j)\text{.}$$


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.