Let us call a series $\sum_n x_n$ in a Banach space "good" if there exists a permutation $\sigma:\mathbb N\to\mathbb N$ such that the rearranged series $\sum_n x_{\sigma(n)}$ converges.

Find a simple proof of the following theorem (which was proved by E.Steinitz in 1913 according to V.Kadets).

Theorem. A series $\sum_n x_n$ in a finite-dimensional Banach space $X$ is "good" if and only if for every linear function $f:X\to\mathbb R$ the series $\sum_n f(x_n)$ is "good".

(This problem was posed 24.09.2017 by Vaja Tarieladze from Tbilisi on page 72 of Volume 1 of the Lviv Scottish Book).

  • $\begingroup$ Well that is obviously false, take for example the function that maps every element of $X$ to zero. Then the sum over the $f(x_n)$ is always good, so it can't be an equivalence. Do you maybe want to consider only injective $f$? $\endgroup$
    – Dirk
    Sep 25 '17 at 10:41
  • $\begingroup$ Oh, I shouldn't read things so fast, sorry... $\endgroup$
    – Dirk
    Sep 26 '17 at 8:48
  • $\begingroup$ With $X=\mathbb{R}^n$, I wonder whether it's enough to take the $f$'s to be the coordinate projections. This would be too good to be true, so does anyone know a counterexample? $\endgroup$ Sep 26 '17 at 14:08
  • 3
    $\begingroup$ @TobiasFritz Let $(x_n)=((1,1)$, $(-1/2,-1/2)$, $(1/3,1/(3\cdot 2))$, $(1/5,1/(3\cdot 2))$, $(-1/4,-1/4)$, $(1/7,1/(5\cdot 4))$, $(1/9,1/(5\cdot 4))$, $(1/11,1/(5\cdot 4))$, $(1/13,1/(5\cdot 4))$, $(-1/6,-1/6)$,$\dots)$. For $k\ge 1$ there's a block of length $2^{k-1}$ where the first coordinate runs through odd fractions and the second coordinate is $1/((2k-1)\cdot 2^{k-1})$, followed by $(-1/(2k),-1/(2k))$. Each coordinate is good, but $f:(a,b)\mapsto a-b$ sends it to a nonnegative series that diverges like $\log(n)-\log\log(n)$. $\endgroup$
    – MTyson
    Sep 26 '17 at 16:18
  • 1
    $\begingroup$ @PietroMajer This problem was resolved by Kadets in (mathnet.ru/links/4cc776c55afef1328f3b1c249757914d/ivm4725.pdf) in case of all Frechet spaces: the Steinitz rearrangement theorem is true for a Frechet space $X$ if and only if $X$ is topologically isomorphic to the countable product of real lines. $\endgroup$ Nov 26 '17 at 14:17

A relatively short inductive proof of Steinitz Theorem can be founded in this paper. Here we present a sketch of the proof, which is based on 3 lemmas whose proof is left to the reader. First we recall the formulation of the result.

Theorem. A series $\sum_{n=1}^\infty x_n$ in a finite-dimensional Banach space $X$ is "good" if and only if for any linear functional $f:X\to\mathbb R$ the series $\sum_{n=1}^\infty f(x_n)$ is "good".

Proof. This theorem will be proved by induction on the dimension of $X$. The theorem is trivially true for Banach spaces of dimension $\le 1$. Assume that for some number $d$ the theorem has been proved for all Banach spaces of dimension $<d$.

Fix a Banach space $X$ of dimension $d$. We can assume that $X=\mathbb R^d$ and the norm $\|\cdot\|$ of $X$ is generated by the standard inner product $\langle\cdot,\cdot\rangle$.

Let $\sum_{n=1}^\infty x_n$ be a series in $X$ such that for every linear functional $f:X\to\mathbb R$ the series $\sum_{n=1}^\infty f(x_n)$ is "good". We lose no generality assuming that $x_n\ne 0$ for all $n$.

Lemma 1. $\lim_{n\to\infty}x_n=0$.

Definition. A point $s$ on the unit sphere $S=\{x\in X:\|x\|=1\}$ is called a divergence direction of the series $\sum_{n=1}^\infty x_n$ if for any neighborhood $U\subset S$ of $x$ the set $\mathbb N_U=\{n\in\mathbb N:\frac{x_n}{\|x_n\|}\in U\}$ is infinite and $\sum_{n\in\mathbb N_U}\|x_n\|=\infty$.

Let $D$ be the (closed) set of all divergence directions of $(x_n)_{n\in\mathbb N}$.

If $D=\emptyset$, then by the compactness of the sphere $S$, $\sum_{n=0}^\infty \|x_n\|<\infty$ and the series $\sum_{n=1}^\infty x_n$ is convergent, so is "good".

So, we assume that $D$ is not empty. Using Hahn-Banach Theorem it can be shown that the convex hull $conv(D)$ of $D$ contains zero.

Let $D_0$ be a subset of smallest cardinality in $D$ such that $0\in conv(D_0)$.

It can be shown that the set $D_0$ is affinely independent and zero is contained in the interior of the simplex $conv(D_0)$.

Let $X_0$ be the linear hull of the set $D_0$ and $X_1$ be the orhogonal complement of $X_0$ in the Euclidean space $X$. So, $X=X_0\oplus X_1$.

For $i\in\{0,1\}$ let $pr_i:X\to X_i$ be the orthogonal projections of $X$ onto $X_i$.

Lemma 2. There exists a subset $\Omega\subset \mathbb N$ such that

1) $\sum_{n\in\Omega}\|pr_1(x_n)\|<+\infty$;

2) for any $s\in D$ and a neighborhood $U\subset S$ of $s$ the set $\Omega_U=\{n\in\Omega:\frac{x_n}{\|x_n\|}\in U\}$ is infinite and $\sum\limits_{n\in\Omega_U}\|pr_0(x_n)\|=\infty$.

Lemma 3. There exists a constant $C$ (dependent on $D_0$) such that for any $x\in X_0$, $\varepsilon>0$, and a finite set $\Omega_0\subset \Omega$ there exists a finite set $F\subset \Omega\setminus \Omega_0$ such that

1) $\|x-\sum_{n\in F}pr_0(x_n)\|<\varepsilon$;

2) $\sum_{n\in F}\|x_n\|\le C\max\{\|x\|,\varepsilon\}$.

Since $X_1$ has dimension $<d$, we can apply the inductive assumption and conclude that the series $\sum_{n\in\mathbb N\setminus\Omega}pr_1(x_n)$ is "good", which means that for some permutation $\sigma_1$ of $\mathbb N\setminus\Omega$ the series $\sum_{n\in\mathbb N\setminus\Omega}pr_1(x_{\sigma_1(n)})$ is convergent in the Banach space $X_1$.

Then using Lemma 3, we can extend the permutation $\sigma_1$ of $\mathbb N\setminus\Omega$ to a permutation $\sigma$ of $\mathbb N$ such that $\sum_{n=1}^\infty pr_0(x_{\sigma(n)})=0$.

Since $\sum_{n\in\Omega}\|pr_1(x_n)\|<+\infty$, the series $\sum_{n\in\Omega}pr_1(x_{\sigma(n)})$ converges and then the series $\sum_{n\in\mathbb N}pr_1(x_{\sigma(n)})=\sum_{n\in\mathbb N\setminus\Omega}pr_1(x_{\sigma(n)})+\sum_{n\in\Omega}pr_1(x_{\sigma(n)})$ converges, too.

Then the series $\sum_{n\in\mathbb N}x_{\sigma(n)}=\sum_{n\in\mathbb N}pr_0(x_{\sigma(n)})+pr_1(x_{\sigma(n)})$ also converges and hence the series $\sum_{n\in\mathbb N}x_n$ is "good".


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.