4
$\begingroup$

Consider a (Hausdorff and complete) locally convex topological vector space $V$ and a countable subset $(v_k)_{k=1}^\infty \subset V$ of non-zero vectors.

$(*)$ Under what conditions on this subset are we guaranteed the existence of a sequence of positive real numbers $(\alpha_k)_{k=1}^\infty$, such that the series $\sum_{k=1}^\infty \alpha_k v_k$ converges (that is, the sequence of its partial sums) converges to a vector $v \in V$?

The positivity of the coefficients, $\alpha_k > 0$ for each $k$, is the crucial aspect of the question.

On the one hand, if $V$ is a Banach space, the answer is always Yes. We can always choose $\alpha_k = c_k / \|v_k\|$, where $(c_k)_{k=1}^\infty$ is any summable sequence of positive numbers.

On the other hand, consider the space $c_{00}$ of finite sequences of real numbers, turned into a complete locally convex space in the usual way. The the subset $(v_k)_{k=1}^\infty$, where the $v_k$ is the sequence that is zero everywhere except the $k$-th place, is an example for which such a sequence of $(\alpha_k)_{k=1}^\infty$ does not exist. This is clear because here a necessary and sufficient condition for the convergence of $\sum_{k=1}^\infty \alpha_k v_k$ is that only finitely many $\alpha_k$ are non-zero.

So maybe I could ask the question in a slightly different way. Is there a well-studied (joint?) property of the space $V$ and the subset $(v_k)_{k=1}^\infty$ that is sufficient (hopefully also necessary) for the answer to $(*)$ to be Yes? I was hoping that the answer would be related to one of the standard theorems of functional analysis, but I haven't recognized the right one yet.

$\endgroup$
4
  • $\begingroup$ I know a couple of "usual" ways to make $c_{00}$ into a locally convex space, but unless I have got my definitions completely wrong, they aren't complete... $\endgroup$ Jul 24, 2015 at 18:33
  • $\begingroup$ @NateEldredge: How about the weak* topology on the dual to the Fréchet space of all sequences? $\endgroup$ Jul 24, 2015 at 19:17
  • $\begingroup$ @IgorKhavkine: No, the weak-$*$ topology is not complete, since $c_{00}$ is not the algebraic dual of $s$. You probably mean the locally convex direct sum topology $c_{00} \cong \bigoplus_{n=1}^\infty \mathbb{R}$, which is complete since each of the summands is complete. This is the finest locally convex topology on $c_{00}$. Furthermore, it follows from Köthe, §22.5.(4) that this coincides with both the strong and the Mackey topology. (See also Köthe's section on sequence spaces, §30.2.) $\endgroup$ Mar 21, 2020 at 20:09
  • $\begingroup$ @J.vanDobbendeBruyn You're right. That is how I was thinking about it. Thanks for the correction! $\endgroup$ Mar 22, 2020 at 12:47

2 Answers 2

2
$\begingroup$

As a partial answer, note the following necessary (and often sufficient) criteria.

Let $E$ be a (real or complex) topological vector space. Let us say that a sequence $\{x_n\}_{n=1}^\infty$ in $E$ can be made summable/bounded/to converge to zero if there is a sequence $\{\alpha_n\}_{n=1}^\infty$ of strictly positive real scalars such that $\{\alpha_n x_n\}_{n=1}^\infty$ is summable/bounded/converges to zero.

Proposition. Consider the following statements:

$\text{(a)}$. $\{x_n\}_{n=1}^\infty$ can be made summable;

$\text{(b)}$. $\{x_n\}_{n=1}^\infty$ can be made to converge to zero;

$\text{(c)}$. $\{x_n\}_{n=1}^\infty$ can be made bounded.

In every topological vector space $E$ one has $$\text{(a)} \Longrightarrow \text{(b)} \Longleftrightarrow \text{(c)}.$$ If $E$ is locally convex and sequentially complete, then $$\text{(a)} \Longleftrightarrow \text{(b)} \Longleftrightarrow \text{(c)}. $$

Proof. $\text{(a)} \Longrightarrow \text{(b)}$. Suppose that $\sum_{k=1}^\infty \alpha_k x_k$ converges to $y \in E$. Let $s_n := \sum_{k=1}^n \alpha_k x_k$ denote the $n$-th partial sum. Then $\lim_{n\to\infty} s_n = \lim_{n\to\infty} s_{n+1} = y$, so by continuity of addition $$ \lim_{n\to\infty} \alpha_{n+1} x_{n+1} = \lim_{n\to\infty} s_{n+1} - s_n = y - y = 0. $$

$\text{(b)} \Longrightarrow \text{(c)}$. Every convergent sequence is bounded.

$\text{(c)} \Longrightarrow \text{(b)}$. If $\{\alpha_n x_n\}_{n=1}^\infty$ is bounded, then $\lim_{n\to\infty} \alpha_n \beta_n x_n = 0$ for every sequence $\{\beta_n\}_{n=1}^\infty$ of scalars converging to zero (see e.g. Rudin, Theorem 1.30).

Assume now that $E$ is locally convex and sequentially complete. $\text{(b)} \Longrightarrow \text{(a)}$. Suppose that $\lim_{n\to\infty} \alpha_n x_n = 0$. Define $s_n := \sum_{k=1}^n \frac{\alpha_k}{2^k} x_k$. Use local convexity to prove that $\{s_n\}_{n=1}^\infty$ is Cauchy, and use sequential completeness to conclude that the series $\sum_{k=1}^\infty \frac{\alpha_k}{2^k} x_k$ converges. $\Box$

Some remarks:

  1. Each of the properties $\text{(a)}$, $\text{(b)}$ and $\text{(c)}$ carries over to all weaker (i.e. coarser) topologies, even those with different dual spaces. Interestingly, in locally convex spaces, property $\text{(c)}$ only depends on the dual pair.

  2. If the topology of $E$ is given by a metric $d$, then every sequence $\{x_n\}_{n=1}^\infty$ satisfies $\text{(b)}$. To prove this, note that for every $z \in E$ we have $\lim_{k\to\infty} \frac{1}{k}z = 0$ (by continuity of scalar multiplication). Equivalently: for every $\varepsilon > 0$ there exists some $K_\varepsilon \in \mathbb{N}_1$ so that $d(0,\frac{1}{k} z) < \varepsilon$ for all $k \geq K_\varepsilon$. Using this, we choose a sequence $\{\alpha_n\}_{n=1}^\infty$ of strictly positive real scalars so that $d(0,\alpha_n x_n) < \frac{1}{n}$ for all $n\in\mathbb{N}_1$. Then $\lim_{n\to\infty} \alpha_n x_n = 0$.

  3. The answer of Robert Israel follows from the previous remark and the Proposition.

  4. There are also non-metrizable tvs where every sequence satisfies $\text{(b)}$ (take a metrizable space and pass to a weaker topology). Likewise, there exist non-Fréchet tvs where every sequence satisfies $\text{(a)}$. Interesting follow-up question: if every sequence satisfies $\text{(b)}$ (resp. $\text{(a)}$), then is there necessarily a finer topology which is metrizable (resp. Fréchet)?

$\endgroup$
5
  • 1
    $\begingroup$ Great answer, thanks! And the proof is of course the same if completeness is weakened to sequential completeness. I see now that my example sequence in $c_{00}$ cannot even be made bounded, because every bounded subset of $c_{00}$ must be finite dimensional (Köthe §18.5(4)). This is a useful enough criterion that I'll accept. $\endgroup$ Mar 22, 2020 at 12:46
  • $\begingroup$ @IgorKhavkine: Thanks for the suggestion — I changed it to sequential completeness (and added some additional remarks). I think you provided the perfect example, illustrating exactly what can go wrong, even in the strong dual of a separable Fréchet space. Furthermore, I concur with your retrospective analysis of the example — I think the boundedness condition might be even easier to see than the convergence. $\endgroup$ Mar 22, 2020 at 18:27
  • $\begingroup$ @IgorKhavkine thanks for looking out, but I respectfully disagree with your latest edit. ;-) Maybe my argument could be clearer, but the idea is to fix $n$ and let $k$ go to infinity. The metric provides a way to choose $\alpha_n$ separately for each $n$, but still end up with a convergent sequence $\{\alpha_n x_n\}_{n=1}^\infty$. (Compare with the $\max\{ \lVert \upsilon_j\rVert_k \, : \, k \leq j\}$ construction from Robert Israel's answer — I guess I could do something similar using the concept of F-norms, but the current approach seems easier.) $\endgroup$ Mar 22, 2020 at 22:30
  • $\begingroup$ Update: added a few extra words to make the aforementioned argument clearer. $\endgroup$ Mar 22, 2020 at 23:12
  • $\begingroup$ Sorry, my bad! I got confused by the notation. Thanks for clarifying. $\endgroup$ Mar 23, 2020 at 0:00
2
$\begingroup$

It's true in a Fréchet space. If $\|\cdot\|_k$, $k = 0,1,2,\ldots$, is a family of seminorms that induces the topology, choose $\alpha_j = c_j/\max\{\| v_j \|_k : k \le j\}$ for a summable sequence of positive numbers $c_j$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.