10

I have read that the answer is no, but I am unable to prove it.

Give $C^{\infty}([0,1])$ the metric $$d(f,g) = \sum_{j=0}^{\infty} 2^{-j} \frac{||(f-g)^{(j)}||}{1 + ||(f-g)^{(j)}||}$$ associated to the collection of seminorms $||f^{(j)}||$, the sup-norm of the $j$-th derivative. This makes $C^{\infty}([0,1])$ into a complete metric space.

How do you show that the topological vector space $C^{\infty}([0,1])$ is not a Banach space? That is, there does not exist a norm which induces the same topology.

It does not appear to be enough to show merely that the metric does not arise from a norm, since it is very non-canonical.

Jesse
  • 385

1 Answers1

12

Hints: 1)There is no norm on on $C^{\infty}[0,1]$ which makes the derivative continuous. This is because, $||\frac{d}{dx}(e^{nx})|| \leq c||e^{nx}||$ for all values of $n$ isn't possible.

2) Show that in the above metric the derivative map is continuous.

voldemort
  • 13,182
  • Do you know how to tackle the general case, that is, that there is no norm in $C^\infty([a,b])$, which make it a Banach space? – Tomás Sep 02 '14 at 20:50
  • @Tomás: No I do not, unfortunately. If you know the proof- please paste a link (or give us the argument)- it will be very helpful for me. – voldemort Sep 02 '14 at 20:54
  • 1
    No, I do not know too, however, I think that should be some contradiction, with respect to the derivative operator, if we assumed it to be true. – Tomás Sep 02 '14 at 20:57
  • @Tomás: I also think so- but I have never seen a proof. – voldemort Sep 02 '14 at 20:59
  • 1
    I, too, am also of that opinion, although I, as well, do not have anything with which to back it up. – Jonas Dahlbæk Sep 02 '14 at 21:01
  • @Tomás: why would there be no such norm? I'm pretty sure that any at least $\mathfrak c$-dimensional real (or complex) vector space is isomorphic (as a vector space) to a Hilbert space, so we can give it a Hilbert norm. – tomasz Sep 18 '14 at 19:21
  • You are right @tomasz. Take a look here – Tomás Sep 18 '14 at 20:56
  • 1
    I am sorry, but why doesn't the same argument work for $C^{\infty}[a,b]$ ? – Optional Apr 10 '16 at 16:06
  • @Optional I think you're right. If $\frac{d}{dx}$ is continuous then, you apply it to $e^{nx}$ (well, we must suppose $a<b$ :) and get a contradiction. – Duchamp Gérard H. E. Jul 24 '17 at 19:09