Scribd
Upload
48 views

Hahn-Banach Theorem: Correspondence # 12

This document provides notes on the Hahn-Banach theorem and two of its corollaries. It begins by defining a sublinear functional and proving a lemma about extending linear functionals defined on subspaces while maintaining certain inequalities. The Hahn-Banach theorem is then stated as extending a linear functional defined on a subspace to the whole space while maintaining an inequality with a sublinear functional. The proof considers both separable and non-separable spaces. Two corollaries are stated, with the first corollary's proof shown to extend a continuous linear functional while maintaining its norm. It is noted that the converse of the second corollary does not always hold.

Uploaded by

EDU CIPANA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
48 views

Hahn-Banach Theorem: Correspondence # 12

This document provides notes on the Hahn-Banach theorem and two of its corollaries. It begins by defining a sublinear functional and proving a lemma about extending linear functionals defined on subspaces while maintaining certain inequalities. The Hahn-Banach theorem is then stated as extending a linear functional defined on a subspace to the whole space while maintaining an inequality with a sublinear functional. The proof considers both separable and non-separable spaces. Two corollaries are stated, with the first corollary's proof shown to extend a continuous linear functional while maintaining its norm. It is noted that the converse of the second corollary does not always hold.

Uploaded by

EDU CIPANA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

ECE 580 SPRING 2019

Correspondence # 12 March 3, 2019


Notes On
HAHN-BANACH THEOREM
Extension Form

I start with a definition – that of a sublinear functional.

Definition. Let X be a real vector space (not necessarily normed). A map p : X → R is called a sublinear
functional if it satisfies the following two properties:

p(x + y) ≤ p(x) + p(y) ∀x, y ∈ X (subadditive)


p(αx) = αp(x) ∀α ≥ 0, ∀x ∈ X (positive homogeneity)

Note that “norm” is a sublinear functional.


The next result is a useful property of sublinear functionals.

Lemma. Let M ⊂ X be a subspace, and f a linear functional on M , such that for some sublinear functional
p,
f (x) ≤ p(x) ∀x ∈ M.
Let xo be a fixed element of X. Then, for any real number c, the following are equivalent :

f (x) + λc ≤ p(x + λxo ) ∀x ∈ M, λ ∈ R (1)

−p(−x − xo ) − f (x) ≤ c ≤ p(x + xo ) − f (x) ∀x ∈ M (2)


Furthermore, there is a real number c satisfying (2) and hence (1).
Proof : To show that (1) ⇒ (2), first set λ = 1, and then set λ = −1 and replace x by −x; then (2) is
immediate. To show the converse, first take λ > 0, and replace x by x/λ on the RHS inequality, to obtain
(1). Now take λ < 0, and replace x by x/λ on the LHS inequality, to arrive again at (1). For λ = 0, (1) is
always satisfied, from the definitions of f and p. To produce the desired c, let x, y be arbitrary elements out
of M . Then,
f (x) − f (y) = f (x − y) ≤ p(x − y)
≤ p(x + xo ) + p(−y − xo ) by subadditivity

⇒ −p(−y − xo ) − f (y) ≤ p(x + xo ) − f (x) ∀x, y ∈ M

⇒ sup {−p(−y − xo ) − f (y)} ≤ inf {p(x + xo ) − f (x)}


y∈M x∈M

Any c between the “sup” and “inf” values will do it. ⋄


Now we have the main theorem.

Theorem (Hahn-Banach). Given a real linear vector space X and a subspace M , let p be a sublinear
functional on X, and f a linear functional on M such that f (x) ≤ p(x) ∀x ∈ M . Then, there is a linear
functional F on X such that F = f on M and F ≤ p on all of X.
Proof. Let x0 ̸∈ M, x0 ∈ X, and consider the subspace M0 consisting of all elements x + λx0 , x ∈ M, λ ∈ R.
We may extend f to a linear functional on M0 by defining

f0 (x + λx0 ) = f (x) + λc

1
where c is any real number. Now choose c such that it satisfies condition (2) in the previous Lemma. Then,
f0 (x + λx0 ) ≤ p(x + λx0 ), ⇐⇒ f0 ≤ p on M0
Hence, we have seen how to extend f from M to M0 , which is of dimension one higher than that of M
(assuming that M is finite-dimensional).
1. If X is normed and separable, then we can find a countable set of vectors {x0 , x1 , . . .}, all linearly
independent and not in M , so that every element in X can be approximated to any degree by linear
combinations of vectors out of M and {x0 , x1 , . . .}. Let Mi = Mi−1 + λxi , which is a subspace. If
y ∈ Mi , we can write it as y = x + λxi , where x ∈ Mi−1 and λ ∈ R. Define
fi (x + λxi ) = fi−1 (x) + λci
where ci is as in the Lemma with f replaced by fi−1 . This then shows how we can extend f recursively
to a countable dense subset of X. Call this extension g, which is naturally linear. By construction,
g ≤ p ∀x ∈ S, where S is a dense subset of X. Now, to extend g to X, let x ∈ X be given and {sn }
be a sequence out of S converging to x. Then, the limit limn→∞ g(sn ) is well defined; call this F (x).
This defines F pointwise. F is obviously linear. Furthermore, since g(sn ) ≤ p(sn ) ∀sn ∈ S, we have
F (x) ≤ p(x) ∀x ∈ X .

2. If X is not a separable normed space, we cannot find an ordered set of vectors to form a basis for a
dense subset. Then, the proof will have to be modified, where the strict ordering is replaced by partial
ordering, and Zorn’s Lemma is used. Zorn’s Lemma says that: “If P is a partially ordered set in which
every chain has an upper bound, then P possesses a maximal element.” This is equivalent to the Axiom
of Choice, which says the following: “Given a nonempty class of disjoint nonempty sets, a set can be
formed which contains precisely one element taken from each set in the given class.” To use this in our
proof, let C be the collection of all pairs (h, H) where h is an extension of f to the subspace H ⊃ M ,
and h ≤ p on H. Partially order C by (h1 , H1 ) ≤ (h2 , H2 ) iff H1 ⊂ H2 and h1 = h2 on H1 . Then, every
chain in C has an upper bound (consider the union of all subspaces in the chain). This implies by Zorn’s
Lemma that C has a maximal element. Call this (F, F). If F ̸= X, then this means that we can extend
F to a larger subspace – but this contradicts its maximality. Hence F is defined on X. ⋄
The two corollaries below follow from the Hahn-Banach theorem. I provide a proof for only the first one.
Corollary 1. Let f be a continuous linear functional defined on a subspace M of X, a normed linear space.
Then, there exists F ∈ X ∗ such that ∥f ∥M = ∥F ∥X and F is an extension of f .
Proof. Take p(x) = ∥f ∥M ∥x∥X in the Hahn-Banach Theorem, which is clearly a sublinear functional. Now
note that
F (x) ≤ p(x) = ∥f ∥M ∥x∥X and − F (x) = F (−x) ≤ ∥f ∥M ∥x∥X ∀ x ∈ X ,
which leads to
|F (x)| ≤ ∥f ∥M ∥x∥X ⇒ sup |F (x)| ≤ ∥f ∥M .
x∈X,∥x∥≤1

But since sup |F (x)| ≥ sup |F (x)| = ∥f ∥M , we actually have equality above. Since F is bounded,
x∈X,∥x∥≤1 x∈M,∥x∥≤1
it is also continuous. ⋄
Corollary 2. Let xo ∈ X, a normed linear space. Then, there exists a nonzero bounded linear functional
F on X such that
F (xo ) = ∥F ∥∥xo ∥ .

The converse of this last result does not generally hold; that is, given F ∈ X ∗ , we may not be able to
find x ∈ X such that F (x) = ∥F ∥∥x∥. See, for example, Example 1 on page 113 of the text by Luenberger.
We will see (in class) that for some (but not all) Banach spaces this converse indeed holds.

♢ ♢ ♢

You might also like