20

I know that probably this question has already been answered, but I'd like to present my attempt of solution.

Let $(E,\|\cdot\|)$ be a normed vector space and let $F\subseteq E$ be a finite-dimensional subspace. Suppose that $\dim_{\mathbb{R}} F=n$.

First of all I proved the following lemma.

Lemma 1 Every finite-dimensional normed vector space is complete.

Then, I noticed that $(F,\|\cdot\|_{F})$ is a finite-dimensional normed vector space, where $\|\cdot\|_{F}$ is the induced norm. So, $F$ is complete by Lemma 1.

Lemma 2 If $(E,\|\cdot\|)$ is a normed vector space such that $\dim_{\mathbb{R}}{E}=n$, then $E$ is algebraically and topologically isomorphic to $\mathbb{R}^{n}$.

By Lemma 2, I can view $F$ as a complete subspace of $\mathbb{R}^n$ and, since $\mathbb{R}^n$ is itself complete, I can conclude that $F$ is closed. Here I used the fact that every complete subspace of a complete space is closed.

Is there something wrong?

davyjones
  • 641
avati91
  • 2,877

1 Answers1

14

The part with lemma 2 isn't necessary. Any complete subspace of a metric space is closed. So the proposition follows directly from lemma 1 and you don't need to assume that $E$ is finite-dimensional. But can you prove lemma 1?

Actually even if you use just “complete subspace of complete space is closed” you don't need lemma 2 since both $E$, $F$ are complete by lemma 1.

On the other hand lemma 1 is a consequence of lemma 2. That's because a homeomorphism between normed linear spaces which is also linear isomorphism preserves completeness. It is so because the mapping is then bilipschitz which goes from the following fact:

For a linear map between normed linear spaces the following are equivalent:

  1. it is bounded
  2. it is lipschitz
  3. it is continuous
  4. it is continuous at zero

The only non-trivial implication is $(4) \implies (1)$. By continuity at zero, there is $δ > 0$ such that $f[B(0, δ)] ⊆ B(f(0), 1)$. But then $f[B(0, 1)] ⊆ B(f(0), \frac{1}{δ})$ and $\lVert f(x)\rVert ≤ \frac{1}{δ}\lVert x \rVert$ for every $x$ by linearity.

user87690
  • 9,133
  • Why is its consequence? Does any linear and topological isomorphism preserve completeness? – avati91 Jul 02 '14 at 12:44
  • @avati91 Yes, you are right. It preserves just complete metrizability. But maybe it can be strengthened. However there is no point in proving completeness of $F$ by lemma 1 while proving completeness of $E$ by lemma 2, which is incorrect as you said. – user87690 Jul 02 '14 at 12:49
  • 1
    @avati91 Actually a homeomorphism between two normed spaces which is linear isomorphism preserves completeness, because it is both lipschitz and inverse lipschitz. – user87690 Jul 02 '14 at 12:55
  • @avati91 Yes, it holds. – user87690 Jul 02 '14 at 13:01
  • If $T:(E,|\cdot|_E)\to (F,|\cdot|_F)$ is a linear and topological isomorphism, then it is also a bi-Lipschitz map. But...why? Which is the Lipschitz constant? – avati91 Jul 02 '14 at 13:03
  • @avati91 If the inverse is continuous, then its bounded (i.e. Lipschitz)...right? In fact, you just need a continuous linear isomorphism by the open mapping theorem. – dessin d'enfant terrible Jul 02 '14 at 13:08
  • I don't agree with you. This theorem, a corollary of the Open Mapping Theorem, applies only if you have to do with Banach Spaces... – avati91 Jul 02 '14 at 13:11
  • @avati91 you don't need the open mapping theorem if you're already assuming a homeomorphism. I was just trying to give a stronger statement. user8790's answer looks right. – dessin d'enfant terrible Jul 02 '14 at 13:28
  • Let $T:(E,|\cdot|_E)\to (F,|\cdot|_F)$ be an algebraic and topological isomorphism. I have to find $L>0$ such that for every $x\in E$ $1/L|x|_E\le |Tx|_F\le L|x|_E$ – avati91 Jul 02 '14 at 13:31
  • @avati91 that's correct. Do you see why, if T is a linear homeomorphism, then $T$ and $T^{-1}$ are bounded operators? – dessin d'enfant terrible Jul 02 '14 at 13:36
  • Is it true that $\frac{1}{|T^{-1}|}=|T|$? – avati91 Jul 02 '14 at 13:43
  • @studiosus I don't think so, but once you admit $||T^{-1}||<\infty$, then choose some $L > max { ||T||, ||T^{-1}|| }$. Then $||T^{-1}(Tx)|| \leq L ||T(x)|| \leq L^{2} ||x||$. Now divide by $L$. – dessin d'enfant terrible Jul 02 '14 at 13:48