3

I know that if $H$ is a Hilbert space and $C$ a closed subspace one can define the orthogonal projection onto $C$ as the map $x \oplus y \in H = C \oplus C^\bot \mapsto x$.

I am wondering:

Is it valid, for an open subspace $O$, to define a (possibly non-orthogonal) projection onto $O$?

And if the answer is affirmative, how to do it?

  • 2
    Careful! "Open" and "non-closed" are very different things. A Hilbert space has only one open subspace, and it happens to also be closed... – Nate Eldredge Feb 15 '15 at 08:33
  • I tried to prove the assertion but never really got to use that C is closed, so now I'm wondering why do I need C to be closed, any thoughts on the matter? thanks a lot – Cristian Baeza Mar 31 '17 at 21:13

1 Answers1

7

If $P$ is a bounded idempotent, i.e., $P^2=P$ (hence it is a projection which is maybe not orthogonal) and its range is $O\subseteq H$, then $O$ is closed. Namely, if $\{ x_n\}_{n=1}^{\infty}$ is a sequence in $O$ which converges to $x\in H$, then $$ \| Px-x\|\leq \| Px-Px_n\|+\|x_n-x\|\leq (1+\| P\|)\| x-x_n\|\to 0$$ which means $Px=x$, i.e., $x\in O$.

Janko Bracic
  • 3,013