Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Given a Hilbert space $H$ of finite dimension, why is any subspace of this space closed? I tried bashing out an answer using an arbitrary Cauchy sequence $\{ f_1 , f_2, \ldots \} \subset S \subset H $ and trying to show its limit $f \in S$. I keep getting stuck and suspect there's an easy answer that I'm missing. Could someone enlighten me on this? Thanks in advance!

share|cite|improve this question
It is interesting to know every subspace of such space are closed. – Paul Aug 2 '12 at 7:23
More generally, every finite-dimensional subspace of a linear normed space is closed, see PlanetMath or elsewhere. – Martin Sleziak Aug 2 '12 at 8:49
up vote 4 down vote accepted

Let $S$ a subspace of $H$, and $\{e_1,\dots,e_d\}$ an orthonormal basis of $S$. We can complete it as a basis of $H$. By Gram-Schmidt process, we can assume that this gives an orthonormal basis $\{e_1,\dots,e_d,f_1,\dots,f_N\}$ of $H$. Then we notice that $S=\operatorname{Span}(f_j,1\leq j\leq N)^{\perp}$, and the orthogonal of a set is closed.

share|cite|improve this answer
Very nice answer! – Rudy the Reindeer Aug 2 '12 at 9:13

Let $H$ be a finite dimensional Hilbert space and $V$ a subspace of $H$. If $V=0$, it's obvious that $V$ is closed. Suppose $V \ne 0$, and let $v_1,\ldots, v_m$ be an orthonormal basis of $V$, with $1 \le m \le \dim H$. Let $x \in \overline{V}$ and $(x^n)_n \subset V$ a convergent sequence whose limit is $x$. Thanks to the Cauchy-Schwarz inequality we have for every $1 \le k \le m$: $$ \left|\langle v_k,x\rangle_H-\langle v_k,x_n\rangle_H\right|=\left|\langle v_k,x-x_n\rangle_H\right| \le \|v_k\|_H\|x-x_n\|_H=\|x-x_n\|_H. $$ Hence $$ \lim_n\langle v_k,x_n\rangle_H=\langle v_k,x\rangle_H \quad \forall\ 1 \le k \le m. $$ It follows that $$ x=\lim_n x_n=\lim_n\sum_{k=1}^m\langle v_k,x_n\rangle_Hv_k=\sum_{k=1}^m\langle v_k,x\rangle_Hv_k, $$ i.e. $x \in V$. Thus $\overline{V} \subset V$, and $V$ is closed.

share|cite|improve this answer
Notice your argument is better than the previous one as it doesn't use that $H$ is finite dimensional. – chango Nov 3 '12 at 10:49
Actually yes it does. How else could we have chosen the basis $v_1, \dots, v_m$? – Joshua Ciappara May 29 '13 at 1:40

A subspace of a finite dimensional vector space is always a finite intersection of hyperplanes.

Under the Hilbert space topology hyperplanes are closed (in fact they are the zero sets of linear forms).

share|cite|improve this answer

I think one can also argue as follows:

(i) Complete subsets of complete metric spaces are closed.

(ii) Every finite dimensional normed space is complete (see here for proof)

$S$ and $H$ are finite dimensional hence by (i) and (ii), $S$ is closed.

share|cite|improve this answer
Did you want to write complete metric space instead of complete topological space? (I might be wrong, but I think that you need at least uniformity to speak about completeness.) This is proved at PlanetMath, too. – Martin Sleziak Aug 2 '12 at 9:17
@MartinSleziak Thank you for the comment. I think the easiest way to fix it is, yes, to replace "topological" with "metric" or "normed". But one can also define Cauchy sequences in spaces with countable neighbourhood bases. – Rudy the Reindeer Aug 2 '12 at 9:20
No, a uniformity is not a topology (but it gives one). It's a uniform structure - a "system of neighborhoods" of the diagonal (Martin mentioned that word already, otherwise I wouldn't have). A topological space admits a uniformity iff it's $T_{3 \,1/2}$. In a metric space the uniform structure is generated by pairs of points such that $d(x,y) \lt \varepsilon$. In a topological (abelian) group you get a uniformity from a countable neighborhood base $\{U_n\}_{n=1}^\infty$ by considering the pairs of points $x,y$ such that $x-y \in U_n$. – t.b. Aug 2 '12 at 9:47
@t.b. Ok. But I think my answer is correct now. (i) is true and a Hilbert space of course is also a metric space. No? – Rudy the Reindeer Aug 2 '12 at 9:51
I was only referring to your first comment. Your answer is correct. – t.b. Aug 2 '12 at 9:51

While I personally prefer Davided's answer, let me show you another more crude way to do it.

Fix an orthonormal basis $e_1,\ldots,e_n$. Each element in your Cauchy sequence is then $$ f_j=\sum_k f_{kj}e_k, $$ for numbers $f_{kj}$. As $\|f_j-f_i\|^2=\sum_k|f_{kj}-f_{ki}|^2$, it is easy to see that each sequence (of numbers) $\{f_{kj}\}_j$ is Cauchy, $k=1,\ldots,n$.

Now you can take convergent subsequences one by one, as we only have $n$ sequences, and so there exist numbers $f_{k,0}$ with $f_{kj}\to f_{k,0}$. It is easy to see then that $$ f_j\to\sum_kf_{k,0}e_k $$ in $H$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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