4
$\begingroup$

Let $V$ be a vector space over $\mathbb{R}$.

Let $M \subset V$ (as a set), and let $d:M \times M \rightarrow \mathbb{R}$ be such that $(M,d)$ is a complete metric space.

Let $W$ be a vector subspace of $V$, and assume that $W \subset M$. Moreover, let $|| \cdot ||: W \rightarrow \mathbb{R}$ be a norm on $W$, and assume that it coincides with $d$; i.e. for any $u,v \in W$, we have $d(u,v) = ||u-v||$.

So, we have a normed vector space $W$ residing in a complete metric space $M$ contained in a vector space $V$.

Let $\overline{W}$ denote the closure of $W$ in $M$. My question is the following: is $\overline{W}$ closed under addition (where addition takes place in $V$)?

My thought process is that any element $x$ of $\overline{W}$ can be described as the limit of a Cauchy sequence $(x_n)_n$ of elements in $W$. Then if we have $x,y \in \overline{W}$ and $(x_n)_n, (y_n)_n \subset W$ with $x_n \rightarrow x$ and $y_n \rightarrow y$, then the sequence $(x_n + y_n)_n$ is Cauchy. By completeness of $M$, it converges to some element $s$. However, I don't know how to show $s = x+y$, or if this is even true.

I suspect, in fact, that the result is not true. If so, I would like to see a counter-example, and possibly extra restrictions that could make it true.

EDIT: As an example, consider $V$ to be the vector space of functions $\{f:\mathbb{R} \rightarrow \mathbb{R}\}$, $M = L^1(\mathbb{R})$ with the metric induced by the norm, and $W$ be the set of step functions with finite support, with the $L^1$ norm.

$\endgroup$
6
  • 1
    $\begingroup$ @J.R. I am asking that on $W$, the metric be equal to the one induced by the norm. I think that is a way of making the metric compatible with the vector space structure. $\endgroup$
    – Sambo
    May 25 '18 at 21:31
  • $\begingroup$ Could it be proved such a $\;W\;$ is closed or, at least, complete? $\endgroup$
    – DonAntonio
    May 25 '18 at 21:35
  • 1
    $\begingroup$ My thought is that you would show $\text{lim}_{n\rightarrow \infty} x_n+y_n=x+y$ by looking at the limit of $||x+y-(x_n+y_n)||\leq ||x-x_n|| + ||y-y_n||$, where the the right hand side goes to zero by convergence of $x_n$ and $y_n$. Maybe this misses some subtlety which you have thought of already? $\endgroup$
    – mcwiggler
    May 25 '18 at 22:23
  • 1
    $\begingroup$ @mcwiggler Sadly, it does. There was actually a (now-deleted) answer with the same mistake: you can't have $x+y$ inside the norm, since you don't know that it belongs to $M$. $\endgroup$
    – Sambo
    May 25 '18 at 22:25
  • 1
    $\begingroup$ @Sambo Of course, very obvious once it's mentioned! (: I will keep puzzling... $\endgroup$
    – mcwiggler
    May 25 '18 at 22:27
3
$\begingroup$

The assumptions don't seem to require any relation between the metric on $M\setminus W$ with the vector space structure on $V$, so it would seem we can fabricate a counterexample by brute force:

Let $W$ be your favorite non-complete normed vector space -- such as, for concreteness, the space of real sequences with finite support under the sup norm.

Let $M$ be the metric completion of $W$.

Let $V$ be the direct sum of $W$ and the space of finite formal linear combinations of elements of $M\setminus W$. ($M$ embeds into this in the obvious way...)

Then, obviously, $\overline W=M$, but $M$ is not closed under addition.

$\endgroup$
6
  • $\begingroup$ Hmm, this is actually identical to Rhys Steele's earlier answer, just terser ... $\endgroup$ May 25 '18 at 23:28
  • $\begingroup$ Stupid question: how do we know that $M \neq V$? $\endgroup$
    – user211599
    May 25 '18 at 23:30
  • $\begingroup$ What is the metric completion of $c_{00}$? Is it not the same as its norm completion $c_0$, which is indeed a vector space? $\endgroup$ May 25 '18 at 23:39
  • 1
    $\begingroup$ @mechanodroid $c_0$ is a completion of $c_{00}$ but it doesnt remain a vector space when you give it the additive structure described here. The formal linear combination of $x$ and $y$ denoted by $x+y$ is not the same thing as their sum in $c_{0}$. $\endgroup$ May 25 '18 at 23:47
  • 1
    $\begingroup$ @JefferyOpoku-Mensah the space of finite formal linear combinations of elements of $M \setminus W$ (the free vector space over $M \setminus W$) is strictly larger than $M \setminus W$. $\endgroup$ May 25 '18 at 23:49
2
$\begingroup$

Let $W$ be a subspace of a Banach space $M$ such that $W$ is not closed in $M$ and let the metric on $M$ be the one induced by its norm. Concretely, take e.g. $W$ to be the space of polynomials on $\mathbb{R}$ and $M$ to be the space of continuous functions with the $\sup$-norm.

In particular $M$ is a complete metric space and the restriction of the metric to $W$ coincides with the norm on $W$.

My idea is to enlarge $M$ and change the additive structure on $M \setminus W$ in a way that is suitably incompatible with the original addition there.

Let $F$ be the free vector space over the set $M \setminus W$. Now let $V = F \oplus W$ be the vector space direct sum of $F$ and $W$. I make the obvious identification of $W$ with $\tilde{W} = \{0\} \oplus W$ as a vector space and of $M$ with $\tilde{M} = [(M \setminus W) \oplus \{0\}] \cup [\{0\} \oplus W]$ as a metric space.

Notice that the vector space structure induced on $\tilde{W}$ by $V$ is the same as its original structure. From here on, all topological operations take place in the topology induced by the metric on $\tilde{M}$ and vector space operations take place in $V$.

Finally, take $x_n, y_n \in W$ such that $x_n \to x \in M \setminus W$, $y_n \to y \in M \setminus W$. Translating this all in to $V$ we have $(0, x_n) \to (x,0) \in (M \setminus W) \oplus \{0\}$ and $(0,y_n) \to (y,0) \in (M \setminus W) \oplus \{0\}$.

However $(x,0) + (y,0) = (x+y,0) \not \in \tilde{M}$ since the addition takes place in the free vector space over $M \setminus W$ so that $x+y \not \in M \setminus W$. Obviously, $\overline{\{0\} \oplus W} \subset \tilde{M}$ and $(x,0), (y,0) \in \overline{\{0\} \oplus W}$ so we see that the closure isn't closed under addition in $V$.

$\endgroup$

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.