02.08.2013 Views

Banach Lattices and the Dual of C(X)

Banach Lattices and the Dual of C(X)

Banach Lattices and the Dual of C(X)

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Ma 110a B. Simon<br />

Fall 2006<br />

GENERAL LINEAR FUNCTIONALS ON C(X),<br />

HAHN AND JORDAN DECOMPOSITIONS<br />

c○ 2006 by Barry Simon<br />

1. <strong>Banach</strong> <strong>Lattices</strong><br />

A lattice is a partially ordered set, L, (i.e., a ≤ b <strong>and</strong> b ≤ c ⇒ a ≤ c<br />

<strong>and</strong> a ≤ b plus b ≤ a ⇔ a = b) so that every a, b have a greatest lower<br />

bound (i.e., c ≤ a, c ≤ b so that d ≤ a plus d ≤ b ⇒ d ≤ c) denoted<br />

a ∧ b, <strong>and</strong> a least upper bound (analog <strong>of</strong> glb with ≤ replaced by ≥),<br />

denoted a ∨ b. a ∧ b is also called min(a, b) <strong>and</strong> a ∨ b is max(a, b).<br />

A vector lattice or Riesz space, V, is a real vector space with a partial<br />

order in which it is a lattice <strong>and</strong> so that for any x ∈ V, a ≤ b ⇒ x+a ≤<br />

x + b <strong>and</strong> any λ ∈ (0, ∞), a ≤ b ⇒ λa ≤ λb.<br />

Clearly, x+(a∧b) = (x+a)∧(x+b) <strong>and</strong> x+(a∨b) = (x+a)∨(x+b).<br />

Also, adding −y − x to y ≤ x shows<br />

Thus, multiplication by −1 reverses order, so<br />

x ≤ y ⇔ −y ≤ −x (1.1)<br />

x ∧ y = −[(−x) ∨ (−y)]<br />

Since (x + y) + [(−x) ∨ (−y)] = y ∨ x, this implies<br />

If V is a vector lattice, define<br />

x ∧ y + y ∨ x = x + y (1.2)<br />

V+ = {x | x ≥ 0} V− = {x | −x ≥ 0}<br />

V+ is called <strong>the</strong> positive elements. Clearly, by (1.1), V− = {x | x ≤ 0},<br />

so <strong>the</strong> partial ordering property shows<br />

Given a vector lattice, V, define<br />

V+ ∩ V− = {0}<br />

x+ = x ∨ 0 x− = (−x) ∨ 0<br />

Each is clearly a positive element. Moreover,<br />

x+ − x = −x + (x ∨ 0) = 0 ∨ (−x) = x−<br />

1


2 NOTES FOR MA 110A<br />

so x = x+ − x−. We define |x| by<br />

|x| = x+ + x− x = x+ − x− (1.3)<br />

A <strong>Banach</strong> lattice is a vector lattice, X, which is a <strong>Banach</strong> space <strong>and</strong><br />

which obeys<br />

|x| ≤ |y| ⇒ x ≤ y (1.4)<br />

for any x ∈ X. Notice since | |x| | = |x|, (1.4) implies<br />

|x| = x (1.5)<br />

Examples. If X is a compact Hausdorff space, C(X) is a <strong>Banach</strong><br />

lattice with pointwise order (i.e., f ≤ g ⇔ f(x) ≤ g(x) for all x) <strong>and</strong><br />

· ∞. Similarly, L p (X, dµ) is a <strong>Banach</strong> lattice.<br />

We need <strong>the</strong> following:<br />

Lemma. Let V be a vector lattice.<br />

(i) If x = a+ − a− with a± ≥ 0, <strong>the</strong>n x± ≤ a±.<br />

(ii) |x| = x+ ∨ x−.<br />

Pro<strong>of</strong>. (i) Since x + a− = a+, we have a+ ≥ x, <strong>and</strong> clearly, a+ ≥ 0, so<br />

a+ ≥ x ∨ 0 = x+. Similarly, a− ≥ (−x) ∨ 0 = x−.<br />

(ii) By (1.2), this is equivalent to x+ ∧x− = 0. Since 0 ≤ x+, 0 ≤ x−,<br />

y = x+ ∧ x− ≥ 0. We have x = (x+ − y) − (x− − y) is a difference <strong>of</strong><br />

positive elements since x± − y ≥ 0, so by (i), y ≤ 0. Thus, y = 0. <br />

Proposition. We have that in any vector lattice,<br />

−g ≤ f ≤ g ⇒ |f| ≤ g (1.6)<br />

Pro<strong>of</strong>. First, −g ≤ g ⇒ 2g ≥ 0 ⇒ g ≥ 0. Next, f ≤ g <strong>and</strong> g ≥ 0<br />

implies f+ ≤ g. Similarly, f− ≤ g. By <strong>the</strong> lemma,<br />

|f| = f+ ∨ f− ≤ g<br />

2. Decomposition <strong>of</strong> BLFs on <strong>Banach</strong> <strong>Lattices</strong><br />

The main result here is<br />

Theorem. Let X be a <strong>Banach</strong> lattice. Any bounded linear function<br />

ℓ: X → R can be written<br />

ℓ = ℓ+ − ℓ−<br />

<br />

(2.1)<br />

where ℓ± ≥ 0 (i.e., f ≥ 0 ⇒ ℓ±(f) ≥ 0). Moreover, <strong>the</strong> decomposition<br />

can be done so that<br />

(i) If ℓ = k+ − k− with k± ≥ 0, <strong>the</strong>n ℓ± ≤ k±.


(ii) For any f ≥ 0, we have<br />

NOTES FOR MA 110A 3<br />

ℓ+(f) + ℓ−(f) ≤ ℓ f (2.2)<br />

Remark. It is an interesting exercise to show this implies X ∗ with<br />

<strong>the</strong> order ℓ ≥ k if <strong>and</strong> only if ℓ(f) ≥ k(f) for f ≥ 0 is a lattice <strong>and</strong><br />

ℓ± = 0 ∨ (±ℓ).<br />

The interest for us will be what it says about C(X) ∗ .<br />

Lemma. In a vector lattice, if 0 ≤ f ≤ g + h with g ≥ 0, h ≥ 0, <strong>the</strong>n<br />

we can find g1, h1 so 0 ≤ g1 ≤ g, 0 ≤ h1 ≤ h, <strong>and</strong> f = g1 + h1.<br />

Pro<strong>of</strong>. Let g1 = f ∧ g so 0 ≤ g1 ≤ g. Define h1 = f − g1 ≥ 0 since<br />

g1 ≤ f. Moreover,<br />

g1 + h = (f + h) ∧ (g + h) ≥ f<br />

since f + h ≥ f <strong>and</strong> g + h ≥ f. Thus, h1 = f − g1 ≤ h. <br />

Pro<strong>of</strong> <strong>of</strong> Theorem. Define for f ≥ 0,<br />

ℓ+(f) ≡ sup(ℓ(g) | 0 ≤ g ≤ f)<br />

Clearly (taking g = 0), ℓ+(f) ≥ 0 <strong>and</strong><br />

so ℓ+(f) < ∞ <strong>and</strong><br />

for<br />

We claim for g, h ≥ 0,<br />

|ℓ(g)| ≤ ℓ g ≤ ℓ f<br />

ℓ+(f) ≤ ℓ f<br />

ℓ+(g + h) = sup(ℓ(f) | 0 ≤ f ≤ g + h)<br />

ℓ+(g + h) = ℓ+(g) + ℓ+(h) (2.3)<br />

= sup(ℓ(g1 + h1) | 0 ≤ g1 ≤ g, 0 ≤ h1 ≤ h) (by <strong>the</strong> lemma)<br />

= ℓ+(g) + ℓ+(h)<br />

Moreover, if λ > 0 is real, ℓ+(λf) = λℓ+(f).<br />

Now define for any f,<br />

ℓ+(f) = ℓ+(f+) − ℓ+(f−)<br />

Then, if<br />

f = q+ − q−<br />

with q+, q− ≥ 0, q+ + f− = q− + f+, so by (2.3),<br />

In particular, since<br />

ℓ+(f) = ℓ+(q+) − ℓ+(q−)<br />

g + h = (g+ + h+) − (g− + h−)


4 NOTES FOR MA 110A<br />

we find that<br />

For λ > 0 <strong>and</strong> real,<br />

<strong>and</strong><br />

ℓ+(g + h) = ℓ+(g+ + h+) − ℓ+(g− + h−)<br />

= ℓ+(g+) + ℓ+(h+) − ℓ−(g−) − ℓ+(h−)<br />

= ℓ+(g) + ℓ+(h)<br />

ℓ+(λf) = ℓ+(λf+) − ℓ+(λf−) = λℓ+(f)<br />

ℓ+(−f) = ℓ+(f−) − ℓ+(f+)<br />

= −ℓ+(f)<br />

Therefore, ℓ+ is a linear functional.<br />

Clearly,<br />

so<br />

<strong>and</strong> ℓ = ℓ+ − ℓ−.<br />

Note that<br />

ℓ+(f) ≥ ℓ(f)<br />

ℓ− ≡ ℓ+ − ℓ ≥ 0<br />

ℓ−(f) = sup(ℓ(g − f) | 0 ≤ g ≤ f)<br />

= sup(−ℓ(h) | 0 ≤ h ≤ f) (2.4)<br />

by taking h = f − g.<br />

Suppose ℓ = k+−k−. Then 0 ≤ g ≤ f implies ℓ(g) = k+(g)−k−(g) ≤<br />

k+(g) ≤ k+(f), so taking sups, ℓ+ ≤ k+. Using (2.4), we get ℓ− ≤ k−.<br />

Finally, given f ≥ 0 <strong>and</strong> g, h so 0 ≤ g ≤ f, 0 ≤ h ≤ f, we have<br />

−f ≤ g − h ≤ f, so |g − h| ≤ f <strong>and</strong> so by <strong>the</strong> <strong>Banach</strong> lattice property<br />

g − h ≤ f. Thus,<br />

|ℓ(g) − ℓ(h)| = |ℓ(g − h)|<br />

≤ ℓ g − h<br />

≤ ℓ f<br />

Taking <strong>the</strong> sup <strong>of</strong> ℓ(g) over all g <strong>and</strong> −ℓ(h) over all h, we find that<br />

(2.2) holds. <br />

Corollary. If ℓ ∈ C(X) ∗ , <strong>the</strong>n ℓ = ℓ+ − ℓ− with ℓ± ≥ 0 <strong>and</strong><br />

ℓ+ + ℓ = ℓ (2.5)


NOTES FOR MA 110A 5<br />

Pro<strong>of</strong>. We only need to check (2.5). But ℓ± = ℓ±(1), so<br />

ℓ+ + ℓ− = ℓ+(1) + ℓ−(1) ≤ ℓ<br />

by (2.2) with f = 1. Since ℓ = ℓ+ − ℓ−,<br />

ℓ ≤ ℓ+ + ℓ−<br />

proving (2.5). <br />

A signed Baire measure on X is an assignment <strong>of</strong> a real number,<br />

µ(A), to each Baire set so that <strong>the</strong>re is C < ∞ obeying if X = A1 ∪<br />

· · · ∪ An <strong>and</strong> Ai ∩ Aj = ∅, <strong>the</strong>n<br />

n<br />

|µ(Aj)| ≤ C (2.6)<br />

j=1<br />

<strong>and</strong> for any countable family <strong>of</strong> disjoint sets {Aj} n j=1 , we have µ(∪Aj) =<br />

µ(Aj) (<strong>the</strong> sum converges absolutely by (2.6)). Given such a measure,<br />

as usual we can form a Stieltjes-type functional ℓ on C(X) with<br />

|ℓ(f)| ≤ f∞ so ℓ = ℓ+ − ℓ− <strong>and</strong> µ = µ+ − µ−. We set |µ| = µ+ + µ−.<br />

Conversely, any µ+ − µ− defines a signed measure. We thus have <strong>the</strong><br />

Riesz–Markov <strong>the</strong>orem for signed measures which says <strong>the</strong>re is a oneone<br />

correspondence between signed measures <strong>and</strong> elements <strong>of</strong> C(X) ∗ .<br />

3. Hahn <strong>and</strong> Jordan Decompositions<br />

As we have seen, every signed measure µ induces an ℓ which can<br />

be written as µ+ − µ− where µ± obeys µ+ + µ− = µ <strong>and</strong> if<br />

µ = ν+ − ν− with ν± ≥ 0, <strong>the</strong>n µ± ≤ ν±. Here we will prove two<br />

famous <strong>the</strong>orems:<br />

Hahn Decomposition Theorem. There exists a Baire set, A, so<br />

that for all C,<br />

µ+(C) = µ(A ∩ C) (3.1)<br />

µ−(C) = −µ((X \ A) ∩ C) (3.2)<br />

µ = µ(A) + µ(X \ A) (3.3)<br />

Jordan Decomposition Theorem. µ+ <strong>and</strong> µ− are mutually singular.<br />

Recall |µ| is <strong>the</strong> positive measure µ+ + µ−. We will deduce <strong>the</strong> two<br />

<strong>the</strong>orems above from<br />

Theorem. There exists a Baire function, g, with |g(x)| = 1 for d|µ|<br />

a.e. x so that <br />

f dµ = fg d|µ| (3.4)


6 NOTES FOR MA 110A<br />

for all f ∈ C(X).<br />

Pro<strong>of</strong>. First, without loss <strong>of</strong> generality, we can suppose µ = 1. We<br />

start by noting µ(f+) = µ+(f+) − µ−(f+) ≤ µ+(f+) <strong>and</strong> µ(f+) ≥<br />

−µ−(f+) so |µ(f+)| ≤ |µ|(f+). Similarly, |µ(f−)| ≤ |µ|(f−), so<br />

<br />

|µ(f)| ≤ |f| d|µ| (3.5)<br />

As in von Neumann’s pro<strong>of</strong> <strong>of</strong> <strong>the</strong> RN <strong>the</strong>orem, we have<br />

<br />

|µ(f)| ≤ |f| 2 1/2 d|µ|<br />

since d|µ| = µ = 1, so <strong>the</strong>re is g ∈ L2 with<br />

<br />

µ(f) = g(x)f(x) d|µ|(x) (3.6)<br />

Originally, this is true for f ∈ C(X), but by use <strong>of</strong> <strong>the</strong> dominated<br />

convergence <strong>the</strong>orem, we can define µ(f) for f any bounded Baire<br />

function <strong>and</strong> both (3.6) <strong>and</strong> (3.5) hold.<br />

Now let ε > 0 <strong>and</strong> let f be <strong>the</strong> characteristic function <strong>of</strong> {x | g(x) ><br />

1 + ε} ≡ Aε. Then (3.6) says<br />

<strong>and</strong> (3.5) says<br />

µ(f) ≥ (1 + ε)|µ|(Aε)<br />

|µ(f)| ≤ |µ|(Aε)<br />

It follows that |µ|(Aε) = 0. Taking ε = 1/2 n , we see g(x) ≤ 1 for a.e.<br />

x. Similarly, g(x) ≥ −1, so<br />

so<br />

|g(x)| ≤ 1 a.e. x (3.7)<br />

From (3.6), we have for f ∈ C(X),<br />

<br />

|µ(f)| ≤ |g| d|µ| f∞<br />

(3.8)<br />

<br />

1 = µ ≤ |g| d|µ| (3.9)<br />

Since |µ(X)| = 1 <strong>and</strong> (3.7) holds, we see |g(x)| = 1 for a.e. x. <br />

Pro<strong>of</strong> <strong>of</strong> <strong>the</strong> Hahn Decomposition Theorem. Define<br />

so<br />

A = {x | g(x) = +1}<br />

X \ A = {x | g(x) = −1}


For f ∈ C(X), let<br />

<br />

ν+(f) =<br />

NOTES FOR MA 110A 7<br />

<br />

fχA d|µ| ν−(f) =<br />

fχX\A d|µ|<br />

so ν± ≥ 0 <strong>and</strong> ν+ − ν− = µ. Thus ν± ≥ µ±. But ν+(1) + ν−(1) =<br />

|µ|(X) = 1 = µ+(1) + µ−(1) implies ν+ − µ+ = µ− − ν− = 0. Thus<br />

µ+ is <strong>the</strong> measure associated to χAd|µ|, that is, µ+(B) = |µ|(B ∩ A) =<br />

µ(B ∩ A) since µ−(B ∩ A) = |µ|((X \ A) ∩ (B ∩ A)) = 0.<br />

Pro<strong>of</strong> <strong>of</strong> <strong>the</strong> Jordan Decomposition Theorem. By <strong>the</strong> Hahn <strong>the</strong>orem,<br />

µ+ <strong>and</strong> µ− are mutually singular.<br />

By combining this analysis with <strong>the</strong> von Neumann pro<strong>of</strong>, one gets<br />

Theorem. Let dµ be a signed Baire measure <strong>and</strong> dν a positive Baire<br />

measure. Then<br />

dµ = dµac + dµsing<br />

where<br />

dµac = f(x) dν(x)<br />

for f ∈ L 1 (R, dν) <strong>and</strong> dµsing is singular with respect to dν in <strong>the</strong> sense<br />

that <strong>the</strong>re is A with ν(X \ A) = 0 <strong>and</strong> µ(B) = 0 for all B ≤ A<br />

(equivalently, |µ|(A) = 0).<br />

Pro<strong>of</strong>. We have d|µ| = h dν + d|µ|sing, so dµ = g dν + g d|µ|sing.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!