5
$\begingroup$

If $(a_{n})$ is a conditionally convergent series in real field, then for any real number $\alpha$, there exists a rearrangement $(a_{k_{n}})$ of $(a_{n})$ such that for all even $n$, $a_{k_{n}} \geq 0$, for all odd $n$, $a_{k_{n}} \leq 0$, and $\sum a_{k_{n}} = \alpha$.

This problem boils down to the following problem: can every conditional real series be rearranged to an alternative convergent series. If this is solved, then apply the extensions of Riemann's theorem by Sierpiński, the original problem is done.

I appreciate any suggestion about this problem. Thanks in advance.

$\endgroup$
3
  • $\begingroup$ Without commenting on the content, problems worded as "Show that …" always sound to me like homework, unless there is some discussion of motivation. Where did this problem come from? $\endgroup$
    – LSpice
    Commented Sep 26, 2021 at 4:38
  • 2
    $\begingroup$ @LSpice you are right, this is a challenging problem present in "Supplements to the Exercises in Chapter 1-7 of Walter Rudin's Principles of Mathematical analysis" by Prof. George M. Bergman, on page 41 problem 3.14:6. $\endgroup$
    – hmeng
    Commented Sep 26, 2021 at 4:43
  • $\begingroup$ You need to clarify that the series $\sum_n a_n$ is convergent. So the starting sentence should probably be "If $(a_n)$ is a convergent series which is not absolutely convergent ...". $\endgroup$
    – Kapil
    Commented Sep 26, 2021 at 5:05

2 Answers 2

4
$\begingroup$

Here you prescribe in addition the sequence of signs of the rearranged series in the Riemann-Dini theorem to be alternating, but note that any non-stationary binary sequence of signs does as well. More precisely:

Let $(a_k)_{k\in\mathbb N} $ be an infinitesimal sequence of non-zero real numbers such that $\sum_{k\ge0}a^+_k=\sum_{k\ge0}a^-_k=+\infty$.

Let $\epsilon\in \{-1,1\}^\mathbb{N}$ be a non-eventually constant sequence.

Let $\alpha\in\mathbb R \cup\{ \pm\infty\}$.

Then there exists a permutation $\sigma$ of $\mathbb N$ such that $$\sum_{k=0}^\infty a_{\sigma(k)}=\alpha$$ $$\text{sgn}\,a_{\sigma(k)}=\epsilon_k.$$

To this end: extract a subset $S\subset\mathbb N$ such that $\sum_{k\in S}a_k$ is absolutely summable and $a_k$ are positive resp. negative for infinitely many $k\in S$ (therefore for infinitely many $k\in \mathbb N\setminus S$ as well, because of the assumption $\sum_{k\ge0}a^+_k=\sum_{k\ge0}a^-_k=+\infty$).

Then do the Riemann-Dini bijection $\tau:\mathbb N\to \mathbb N \setminus S$ relatively to the sequence $(a_k)_{k\in \mathbb N\setminus S }$ and the number $\alpha-\sum_{k\in S}a_k$, namely $$\sum_{k=0}^\infty a_{\tau(k)}=\alpha-\sum_{k\in S}a_k.$$

Finally, insert the coefficients $\{a_k\}_{k\in S}$ in some order into the series $\sum_{k=0}^\infty a_{\tau(k)}$, so as to get a rearrangement with the prescribed final sequence of signs $(\dagger)$. Since $\sum_{k\in S}a_k$ is absolutely summable, the order of the insertion does not affect the convergence and the value of the sum, which is $\alpha$ as wanted.

$(\dagger)$ This can easily be done, for the reason that any non eventually constant binary sequence contains any other non eventually constant binary sequence as a subsequence, in such a way that the complement is also non eventually constant.

$\endgroup$
7
  • $\begingroup$ Thanks so much for your generalization of my post, which does really deep my understanding about the conditionally convergent series. However, I still have difficulty to understand the dagger-symbolied sentence. How can we guarantee that the series after the insertion process is still convergent, and moreover, the corresponding convergent result is the sum of the two sub-series. I only know the fact that we can add finite zeros between successive terms of a convergent series still converges to the same value. Any sources (e.g., papers) are welcomed. $\endgroup$
    – hmeng
    Commented Sep 26, 2021 at 16:41
  • $\begingroup$ You're welcome. The $\dagger$ sentence is purely combinatoric. Given two binary sequences $(\epsilon_k)_{k\in\mathbb N}$ and $(\delta_j)_{j\in\mathbb N}$ in $\{-1,+1\}^{\mathbb N}$, both non eventually constant you can certainly find a copy of the latter within the former, that is a subsequence $\epsilon_{k_j}=\delta_j$ for all $j\in \mathbb N$, and you can do so in such a way that between $\epsilon_{k_j}$ and $\epsilon_{k_{j+1}}$ there are at least one $-1$ and one $+1$, so that the "complement" string is not eventually constant. $\endgroup$ Commented Sep 26, 2021 at 20:03
  • $\begingroup$ You apply this to the initially prescribed sequence $\epsilon_k $ and to $\delta_j:=\text{sgn}\,a_{\tau(j)}$. This means that you can fill every place $k_j$ with the coefficient $a_{\tau(j)}$ , that has the prescribed sign $a_{\tau(j)}=\epsilon_{k_j}$ at the place $k_j$. The remaining places $\mathbb N\setminus \{k_j:j\in\mathbb N\}$ can be bijectively filled with the remaining coefficients $\{a_k\}_{k\in S}$ of the prescribed sign, because infinitely many of them must be $+1$ and infinitely many must be $-1$, and you do have enough of both among the $\{a_k\}_{k\in S}$. $\endgroup$ Commented Sep 26, 2021 at 20:15
  • $\begingroup$ Convergence: consider the final rearranged series $\sum_{k=0}^\infty a_{\sigma(k)}$. Any $n$-th partial sum of it is always a $m$-th partial sum of $\sum_{k=0}^\infty a_{\tau(k)}$ plus a $(n-m)$-th partial sum of some fixed rearrangement of $\sum_{k\in S}a_k$, and by construction both $m$ and $n-m$ go to $\infty$ as $n\to\infty$. Since $\sum_{k\in S}a_k$ does not depend on rearrangements because it is absolutely convergent, $\sum_{k=0}^\infty a_{\sigma(k)}$ does converges to $\sum_{k=0}^\infty a_{\tau(k)}+\sum_{k\in S}a_k=\alpha$. $\endgroup$ Commented Sep 26, 2021 at 20:23
  • $\begingroup$ (By means of a more formal notation all the maps $\mathbb{N}\to\mathbb{N}$ I used may be constructed formally, but I thought an informal description could be easier to communicate) $\endgroup$ Commented Sep 26, 2021 at 20:33
1
$\begingroup$

I believe the answer is yes. Let $a_{i_k}$ and $a_{j_k}$ be the even and odd terms respectively, ordered in increasing order of magnitude.

We will define four sequences; $b_n$ (indexed by $\mathbb Z_+$) which will be the desired rearranged sequence, a $-1, 1$ valued sequence $s_n$ (indexed by $\mathbb N$), which will be a “state/control” variable, and two sequences $P_n, M_n$ (indexed by $\mathbb N$) of natural numbers.

To this end, consider the following algorithm:

Start algorithm.

  1. Set $P_0 = M_0 = 0$.

  2. If $|a_{i_0}| \geq |a_{j_0}|$ , set $s_0 = 1$, else set $s_0 = -1$.

  3. Assume $s_0, \dots, s_{n-1}; b_1, \dots, b_{n-1}, P_{n-1}, P_{n-1}$ have been defined,

  4. Set $P_{n}$ (respectively $M_{n})$ to be the largest index $k$ such that $a_{i_m}$ (respectively $a_{j_m}$) has already been used for all $m < k$.

  5. Do the following:

While $s_{n-1} = 1$, and $n$ is odd,

  • If $\overset{n-1}{\underset {i=1} {\sum}} b_r >\alpha - a_{i_{P_{n}}}$, set $b_n = a_{j_{M_n}}$ and set $s_n = 1$; else set $b_n = a_{j_l}$ for any $l \geq M_n$ with $|a_{j_l}| \leq \frac{1}{2}|a_{i_{P_n}}|$ and set $s_n = -1$.

While $s_{n-1} = 1$, and $n$ is even,

  • If $\overset{n-1}{\underset {i=1} {\sum}} b_r >\alpha - a_{i_{P_{n}}}$, set $b_n = a_{i_l}$ for any $l$ with $|a_{i_l}|\leq \frac{1}{2} |a_{M_n}|$ and set $s_n = 1$; else set $b_n = a_{i_{P_n}}$ and set $s_n = -1$.

While $s_{n-1}= -1$, and $n$ is even,

  • If $\overset{n-1}{\underset {i=1} {\sum}} b_r < \alpha - a_{i_{M_{n}}}$, set $b_n = a_{i_{P_n}}$ and set $s_n = -1$; else set $b_n = a_{i_l}$ for any $l \geq P_n$ with $|a_{i_l}| \leq \frac{1}{2}|a_{j_{M_n}}|$ and set $s_n = 1$.

While $s_{n-1} = -1$, and $n$ is odd,

  • If $\overset{n-1}{\underset {i=1} {\sum}} b_r < \alpha - a_{j_{M_{n}}}$, set $b_n = a_{j_l}$ for any $l$ with $|a_{j_l}|\leq \frac{1}{2} |a_{P_n}|$ and set $s_n = -1$; else set $b_n = a_{i_{M_n}}$ and set $s_n = 1$.

End algorithm.

I claim this algorithm exhausts all terms. Indeed it suffices to show that $b_n = a_{i_{P_n}}$ and $b_n = a_{j_{P_n}}$ infinitely often. We prove only the former, the proof for the latter being identical via symmetry.

Thus assume for contradiction that $b_k = a_{i_{P_k}}$ for only finitely many $k$, and let $K$ be the largest of these. Then the terms $b_n$ starting from $n = K+1$ are $a_{i_{M_n}}, a_{i_{l_n}}, a_{i_{M_{n+1}}}, a_{i_{l_{n+1}}}, \dots$, and further it is always the case that $\overset{n-1}{\underset {i=1} {\sum}} b_r >\alpha - a_{i_{P_{n}}}$ for even $n > K$.

But since $|a_{i_{l_n}}| \leq \frac{1}{2}|a_{i_{M_n}}|$, and the sequence $\sum a_{j_k}$ diverges (to negative infinity), we must eventually have $\overset{n-1}{\underset {i=1} {\sum}} b_r < \alpha - a_{i_{P_{n}}}$, contradiction.

Finally, to see that $\sum b_n$ converges to $\alpha$, it suffices to note that by construction, once $s_n$ has switched values at least once, then from that point on $\overset{n}{\underset {i=1} {\sum}} b_r$ is no more than $2(|a_{P_n}| + |a_{M_n}|)$ away from $\alpha$, which goes to 0 as $n \to \infty$.

$\endgroup$
4
  • $\begingroup$ where are $A_{n}$ and $B_{n}$ used? and also seems your $(a_{i_{k}})$ and $(a_{j_{k}})$ are the positive and negative terms in their original order, right? So your comparison at step 2 should be the comparison of their absolute value, right? Sorry I did not go though completely and may misunderstand your points $\endgroup$
    – hmeng
    Commented Sep 26, 2021 at 2:37
  • $\begingroup$ Sorry $A_n$ and $B_n$ should read $P_n$ and $M_n$ respectively. Also yes. The lack of absolute values is a typo. Will correct it. $\endgroup$
    – Nate River
    Commented Sep 26, 2021 at 2:38
  • $\begingroup$ No worries, please allow me some time to digest your process $\endgroup$
    – hmeng
    Commented Sep 26, 2021 at 2:39
  • $\begingroup$ Sure no prob! Btw there is another edit - step 1 should read $P_0 = M_0 = 0$. $\endgroup$
    – Nate River
    Commented Sep 26, 2021 at 3:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .