An infinite series $\Sigma_n a_n$ is said to absolutely converge if $\Sigma_n |a_n|$ converges. Absolute convergence implies convergence.

My question is, is absolute convergence a topological concept? That is to say, does there exist a topology on $\mathbb{R}$ such that $\Sigma_n a_n$ converges in that topology if and only if $\Sigma_n |a_n|$ converges in the standard topology?


Let $\tau$ be a topology on $\mathbb{R}$ under which every absolutely convergent series converges.

Lemma. If $(x_n) \to L$ under the standard topology, then $(x_n) \to L$ under $\tau$.

Proof. Let $(y_n)$ be any subsequence of $(x_n)$. Then there exist $(n_k)$ such that $|y_{n_k} - L| < 2^{-k}$. Now consider the sequence $(a_k)_{k=0}^{\infty}$ defined by

$$ a_k = \begin{cases} L, & k = 0 \\ y_{n_j} - L, & k = 2j-1 \text{ for some } j \geq 1 \\ L - y_{n_j}, & k = 2j \text{ for some } j \geq 1 \end{cases} $$

Then $\sum_{k=0}^{\infty} |a_k| < \infty$, and so its partial sum $s_k = \sum_{j=0}^{k} a_j$ converges under $\tau$. Since $s_{2k} = L$, it follows that $s_k \to L$ under $\tau$. Then, since $s_{2k-1} = y_{n_k}$, we have $y_{n_k} \to L$ under $\tau$.

So far, we have proved that every subsequence of $(x_n)$ has a further subsequence that converges to $L$ under $\tau$. This suffices to guarantee that $(x_n)$ converges to $L$ under $\tau$. ////

By this lemma, any series which converges conditionally under the standard topology also converges under $\tau$.



Choose $c_n\uparrow 0$ and $d_n\downarrow 0$ such that $d_n-c_n=\frac1n$. A series $\sum a_n$ whose partial sums contain both $c_n,d_n$ for all large enough $n$ is obviously not going to converge absolutely, so your topology need to detect this by having a neighbourhood of $0$ that omits infinitely many points from $(c_n)$ or infinitely many from $(d_n)$. Say it is from $c_n$ for definiteness. But then we can construct a series $\sum b_n$ whose partial sums are precisely those omitted points from $(c_n)$. Since $c_n\uparrow 0$, $b_n=c_{s(n)}-c_{s(n-1)}$ are all positive (except the first term which is $c_{s(1)}<0$), so $\sum b_n$ is absolutely convergent, contradiction.

  • $\begingroup$ What if we allow enlarging $\mathbb{R}$ under the new topology? Couldn't $c_n$ converge to something like $0_-$ and $d_n$ to $0_+$? $\endgroup$ – Bananach Nov 9 '18 at 6:42

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.