9
$\begingroup$

On Reed & Simon, vol 2, chapter X, problem 4, it is asked:

Let $M$ and $N$ be closed subspaces of a separable Hilbert space. If $\dim M > \dim N$, prove that $M\cap N^{\perp} \ne \{0\}$.

Here, $\dim V$ is the cardinality of a Hilbert basis for $V$.

The solution is pretty straightforward, since we must have $\dim N = n \in \mathbb{N}$ and therefore it becomes a linear algebra problem.

My question is: does it hold for non-separable Hilbert spaces?

I couldn't prove it or think in counterexamples for it. Thanks in advance!

$\endgroup$
2
  • 4
    $\begingroup$ I think the answer is yes. My sketchy idea is as follows: Consider the orthogonal projection $P(N)$ of $N$ onto $M$ and choose an orthonormal Hilbert basis for this space. Then extend the basis of $P(N)$ to an orthonormal basis for $M$. Since $\dim(P(N))<\dim(M)$, the extension is non-trivial. Any of the new basis vectors in the extension will be in $N^\perp$. I'd need to check with a functional analysis textbook to make sure all the operations described above are legal... $\endgroup$
    – Yly
    May 4 '18 at 5:29
  • $\begingroup$ @Yly. Quite right, And nicely done. $\endgroup$ May 4 '18 at 12:40
5
$\begingroup$

More generally, suppose $M$ and $N$ are Hilbert spaces with $\dim M>\dim N$ and let $T:M\to N$ be a bounded linear map. Then the image of the adjoint map $T^*:N\to M$ cannot be dense in $M$ (the closure of the image is the closed span of the image of an orthonormal basis for $N$, and so has dimension at most $\dim N$). So there exists nonzero $v\in M$ which is orthogonal to the image of $T^*$. This means $0=\langle v,T^*w\rangle=\langle Tv,w\rangle$ for all $w\in N$, so $Tv=0$. That is, the kernel of $T$ is nontrivial.

In your setup, we now immediately conclude that $M\cap N^\perp$ is nontrivial by letting $T:M\to N$ be the orthogonal projection to $N$.

$\endgroup$

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.