An introduction to the Hahn-Banach extension theorem: Part III

In this post we will see how the Hahn-Banach key lemma from Part II allows us to prove the Hahn-Banach theorem for extensions of bounded linear functionals on separable normed spaces, without the need to use Zorn's Lemma (or equivalent).

First, we recall an easy result about extensions of bounded linear functionals to the closure of a linear subspace.

Let F be R or C, and let (E,) be a normed space over F. Now let F be a linear subspace of E. Then the closure of F in E, ¯F, is also a linear subspace of E. As usual, we give F and ¯F the norm obtained by restricting E's norm, .

Proposition 1 (extensions of linear functionals to the closure of a subspace)

Let ψF. Then ψ has a unique extension ˜ψ¯F (the dual space of ¯F), and we have

˜ψop=ψop.

Comments

  • This is a special case of a more general result about extensions of continuous linear maps from F into a Banach space Y. (Of course, (F,||) is a Banach space.)
  • There is only one possible definition we can try for ˜ψ, and it works. For every x in ¯F and every sequence (xn) in F which converges to x, we have to have 
    ˜ψ(x)=limnψ(xn).
    Fortunately you can show that this limit exists, and is independent of which such sequence (xn) in F we take. The rest is fairly routine.
  • We could also note that ψ is Lipschitz continuous and hence uniformly continuous on F. Thus ψ has a unique continuous extension to ¯F, which gives us ˜ψ. Again, the rest is fairly routine.
We are now ready to prove the Hahn-Banach theorem for separable normed spaces over R.

Theorem (Hahn-Banach Extension Theorem for linear functionals on separable normed spaces over R

Let (E,) be a separable normed space over R, and let F be a linear subspace of E. Let ψF. Then there exists ϕE with ϕop=ψop and such that ϕ is an extension of ψ.

Comments
  • As usual, in this theorem and its proof, every linear subspace of E is regarded as a normed space by restricting E's norm, .
  • We denote the linear span of a subset A of E by linA.
  • We start by applying the Hahn-Banach key lemma repeatedly to obtain a sequence of successive extensions of ψ from F to a nested increasing sequence of linear subspaces of E whose union is dense in E
  • From this we obtain an extension of ψ to the relevant union of subspaces, and then apply Proposition 1 above to obtain the extension ϕ in E that we want.
  • In applying the Hahn-Banach key lemma we might appear to be using some form of sequential choice. However, inspection of our proof of the key lemma shows that, once we have specified y there, we can make explicit which constant c we prefer. (For example it could be the minimum constant that works.)
Proof
Since E is separable, there is a countable dense subset S of E. Say 
S={x1,x2,x3,}.
Set F0=F and, for each nN, set
Fn=lin(F{x1,x2,,xn})=F+lin{x1,x2,,xn}.
Note that, for all nN, we have
Fn=Fn1+lin{xn},
and so Fn1 has codimension 1 or 0 in Fn.

We now choose, inductively, ψnFn as follows (for nN{0}).

We set ψ0=ψF=F0.
Now let nN and suppose that we have chosen ψn1Fn1.
  • If xnFn1, then Fn=Fn1 and we take ψn=ψn1Fn.
  • Otherwise xnFnFn1, and Fn=Fn1lin{xn}. In this case we can apply the Hahn-Banach key lemma to obtain ψnFn which extends ψn1, and such that ψnop=ψn1op.
The induction may now proceed. Note:
  • For all nN, we have ψnop=ψop.
  • For all m,nN{0} with mn, ψn is an extension of ψm.
Set 
Fω=n=0Fn=lin(F{x1,x2,})=F+lin{x1,x2,}.
We may define ψωFω as follows: for each nN and each xFn, set
ψω(x)=ψn(x).
  • We need to check that this is well-defined, but fortunately the "consistency" of our extensions above ensures this. If m,nN{0} and x is in both Fm and Fn, then (as noted above), ψm(x)=ψn(x).
  •  It is easy to check that ψω is both linear and continuous, with ψωop=ψop.
  • Clearly Fω is dense in E.
Finally, we apply Proposition 1 above, and let ϕE be the unique continuous linear extension to E of ψω. Then ϕ is an extension of ψ, and 
ϕop=ψωop=ψop, as required.

  • In the proof above we only really needed F+lin{x1,x2,} to be dense in E. For example, this is guaranteed if {x1,x2,} is a Schauder basis of E.
  • In the next post, we see how the general/separable complex Hahn-Banach theorem (for extensions of bounded linear functionals on normed spaces) can be deduced from the corresponding real Hahn-Banach theorem.
  • We will then give a proof of the general (real) Hahn-Banach theorem .
  • In fact we will give two proofs: one the usual proof by Zorn's lemma, another a proof by transfinite induction.
  • You may already be able to see how the transfinite induction argument will go based on the inductive construction above, where we stopped at the first infinite ordinal ω.

Comments

Popular posts from this blog

A rule of thumb for when to use the Ratio Test

An introduction to the Hahn-Banach extension theorem: Part VI

Discussion of the proof that the uniform norm really is a norm