Hahn Banach Theorem states that given a linear continuous functional $f$ on a subspace $N$ of a normed space $M$, it can be extended to a linear functional $F$ on the whole space $M$ and the norm of the extension is the same as the one of $f$.

Can someone tell me if something is wrong in the following lines:

Fact 1: Every vector space has a basis (use axiom of choice if it is infinite dimensional). Fact 2: A linear map is defined if we give the image of a basis.

Take $\{e_i\}_{i\in I}$ a basis of $M$ extending one of $N$, and define $F(e_i):=f(e_i)$ if $e_i$ is in $N$ and $0$ otherwise. This is clearly an extension of $f$, plus the norm of $F$ is the same as that of $f$ because sup $|F(x)|/||x||$ is $0$ for $x \in M \setminus N$ and $||f||$ if $x \in N$.

In case is correct this seems like a very simple proof of this theorem!

  • 2
    $\begingroup$ Hahn Banach Theorem provides a continuous linear extension $F$ of your continuous linear functional $f$. Your "proof" does not take into account continuity of the alleged extension. $\endgroup$ – Marco Vergura Dec 10 '14 at 13:55
  • $\begingroup$ @Marco Vergura Well if $F$ is bounded and linear then it is also continuous. And since $f$ is bounded $F$ it is as well by the last observation $\endgroup$ – inquisitor Dec 10 '14 at 13:59
  • $\begingroup$ @DavidMitra That's even better. I was a little bit too fast in reading the question and I thought the OP had taken a basis of $N$, then completed it to a basis of $M$ and then extended $f$ by zero out of $N$. This process, I guess, tends to make continuity of the extension impossible to control. $\endgroup$ – Marco Vergura Dec 10 '14 at 14:04
  • $\begingroup$ Yes I was thinking to do the extension actually, so I'm going to update the OP. Thanks @David Mitra $\endgroup$ – inquisitor Dec 10 '14 at 14:07
  • 2
    $\begingroup$ Marco's comment is on point. If $x\in M\setminus N$, how do you know that the expansion of $x$ with respect to the basis is a linear combination of $e_i$'s in $M\setminus N$? It seems this is what you're assuming... $\endgroup$ – David Mitra Dec 10 '14 at 14:20

I hope this can serve as a counterexample.

Let $X$ be an infinite dimensional Banach space and $\mathcal{B}=\{e_\gamma: \gamma\in \Gamma \}$ be a Hamel basis of $X$. By the comment of Mike F to this post, we know there exists an unbounded coordinate functional $g_{\gamma_0}$ associated to $e_{\gamma_0}$.

Now let $Y=\text{span}\{ e_{\gamma_0} \}$, a one-dimensional subspace of X. Clearly $g_{\gamma_0}|Y$ is continuous on $Y$ and $g_{\gamma_0}$ is your extension of $g_{\gamma_0}|Y$. But $g_{\gamma_0}$ is unbounded by the first paragraph.


Take the space $\mathbb R^2$ with the usual norm. Choose a non-orthogonal basis. Apply your method to extend from $1$ dimension to $2$. See that the norm is not preserved.


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.