Show that any finite-dimensional subspace $(S,\|\cdot\|)$ of an infinite-dimensional normed vector space $(V,\|\cdot\|)$ is closed and nowhere dense.


  • Let $\{x^{(n)}\}_{n\geq1}$ be a sequence in $S$. Suppose by contradiction that $x^{(n)}\to x \in V\backslash S$, infinite dimensional.
    By the definition of convergence $\|x^{(n)}-x\|\to 0$, but $x=(x_1,x_2,...)$ has at least one coordinate such that $x_i\neq 0= x_i^{(n)}$ for some $i=1,2,...$, where $x_i^{(n)}$ is the $i$-th element of the $n$-th vector of the sequence, and so a contradiction.
  • Not sure what topology should I refer to (product top?!)

Question: can anyboody check the firsts part and give a hint for the second.

  • 2
    $\begingroup$ Why is it a contradiction if some element of $x$ is non zero? What do you mean by coordinate? Does $V$ have some structure in addition to being a normed space? The topology is surely the topology that comes with the associated norm? $\endgroup$ – copper.hat Feb 23 '14 at 0:36
  1. The starting idea to take a sequence $x_n$ in $S$ that converges to some $x$ and see whether $x\in S$ is OK, the rest is nonsense, e.g. nobody said that $x=0$ neither that $x_i^{(n)}=0$, or $S$ may have more than countably infinite dimensions, whence the sequence setting $x=(x_1,x_2,\dots)$ is not that much justified.
    Probably the fastest argument to continue is to observe that $S$ is complete, i.e. is a Banach space (because finite dimensional), and then, since $x_n$ converges to $x\in V$, we have that $x_n$ is a Cauchy sequence, but then it also converges in $S$, so we must have $x\in S$.

  2. The topology is given by the norm: open balls $B_\varepsilon(x):=\{v\in V\mid \|v-x\|<\varepsilon\}$ form a basis of the topology, just as in the finite dimensional case. Hence, every open set is a union of open balls, but any open ball 'uses up' all the dimensions.


Suppose $p \notin S$, and let $v_1,...v_n$ span $S$. Define the functional $f$ on the space spanned by $p,v_1,...v_n$ by $f(v_k) = 0$, $f(p) = 1$. Use Hahn Banach to extend this functional to the whole space. Let $U = \{ x | f(x) > {1 \over 2} \}$ which is open, and since $f(s) = 0$ when $s \in S$, we see that $S \cap U = \emptyset$, hence $S^c$ is open.

Note: The point of the above construction is to create a continuous functional $f$ such that $S \subset \ker f$ and $p \notin \ker f$. Then the set $\{x | f(x) >0 \}= f^{-1}((0,\infty))$ is open (since $f$ is continuous) and $\ker f \cap f^{-1}((0,\infty)) = \emptyset$.

By the same token, we see that if $f(x) \neq 0$, then $x \notin S$.

To see that $S$ is nowhere dense, note that since $S$ is closed, we need only show that it has an empty interior. Suppose $s \in S$, then let $x_n = s+{1 \over n} p$ (where $p$ is from the first paragraph).


Note that $f(x_n) = {1 \over n}$ and so $x_n \notin S$, and we have $\|s-x_n\| = {1 \over n}\|p\|$. Hence $S$ contains no open set.

  • $\begingroup$ Thank you for the very good answer. I though shamefully cannot get the bit "$S \cap U = \emptyset$, hence $S^c$ is open". Is it because somehow if $x\not\in S$ then $x\in U$ by the definition of the linear bdd functional? Could not it be that there is another element $z$ in $S^c$ s.t. $f(z)=0 $? $\endgroup$ – Ton Feb 23 '14 at 16:46
  • 1
    $\begingroup$ Showing that $S$ is closed is equivalent to showing that $S^c$ is open. So I picked a point $p \in S^c$ and created an open set containing $p$ that does not intersect $S$, which shows that $S^c$ is open. It is possible that $\ker f$ contains points other than $S$, but it doesn't matter for the purpose of showing that $S^c$ is open. $\endgroup$ – copper.hat Feb 23 '14 at 17:03

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.