Problem Set

Problem Set #1

Math 5322, Fall 2001

March 4, 2002

ANSWERS
i
 All of the problems are from Chapter 3 of the text.

Problem 1. [Problem 2, page 88]

If ν is a signed measure, E is ν-null iff |ν|(E) = 0. Also, if ν and µ are
signed measures, ν ⊥ µ iff |ν| ⊥ µ iff ν + ⊥ µ and ν − ⊥ µ
Answer :
For the first part of the problem, let X = P ∪ N be a Hahn decomposition of
X with respect to ν. Thus, we have
ν + (E) = ν(E ∩ P )
ν − (E) = −ν(E ∩ N )
|ν| = ν + + ν −
Assume that E is ν-null. This means that for all measurable F ⊆ E, ν(F ) = 0.
But then E ∩ P ⊆ E, so ν + (E) = 0 and E ∩ N ⊆ E, so ν − (E) = 0. Then we
have |ν|(E) = ν + (E) + ν − (E) = 0
Conversely, suppose that |ν|(E) = 0. Let F ⊆ E be measurable. Then
0 ≤ ν + (F ) + ν − (F ) = |ν|(F ) ≤ |ν|(E) = 0, so ν + (F ) = 0 and ν − (F ) = 0. But
then ν(F ) = ν + (F ) − ν − (F ) = 0. Thus, we can conclude that E is ν-null.
For the second part of the problem, we want to show that the following
conditions are equivalent.
(1) ν ⊥ µ
(2) ν + ⊥ µ and ν − ⊥ µ
(3) |ν| ⊥ µ
Let’s first show that (1) =⇒ (2). Since ν ⊥ µ, we can decompose X into a
disjoint union of measurable sets A and B so that A is µ-null and B is ν-null.
We can also find an Hahn decomposition X = P ∪N with respect to ν, as above.
Since B in ν-null, we have ν + (B) = ν(P ∩ B) = 0 and ν − (B) = −ν(B ∩ N ) = 0.
Thus, B is ν + -null and ν − -null. Since A is still µ-null, we conclude that ν + ⊥ µ
and ν − ⊥ µ.
Next, we show that (2) =⇒ (3). Since ν + ⊥ µ, we can find disjoint
measurable sets A1 and B1 so that X = A1 ∪ B1 , ν + (B1 ) = 0 and A1 is µ-null.
Similarly, we can write X = A2 ∪ B2 (disjoint union) where ν − (B2 ) = 0 and A2
is µ-null. We can then partition X as
X = (A1 ∩ A2 ) ∪ (A1 ∩ B2 ) ∪ (A2 ∩ B1 ) ∪ (B1 ∩ B2 )
Now, (A1 ∩ A2 ) is contained in the µ-null set A1 , and so is µ-null. Similarly,
A2 ∩ B2 and A2 ∩ B1 are µ-null. We have 0 ≤ ν + (B1 ∩ B2 ) ≤ ν + (B1 ) = 0, so
ν + (B1 ∩ B2 ) = 0. Similarly, ν − (B1 ∩ B2 ) = 0, and thus |ν|(B1 ∩ B2 ) = 0. We
thus have a disjoint measurable decomposition X = A3 ∪ B3 where
A3 = (A1 ∩ A2 ) ∪ (A1 ∩ B2 ) ∪ (A2 ∩ B1 )
B3 = B1 ∩ B2 .

1
The set A3 is µ-null and B3 is |ν|-null. This shows that |ν| ⊥ µ.
Finally we show that (3) =⇒ (1). So, suppose that |ν| ⊥ µ. The we have
a measurable decomposition X = A ∪ B where |ν|(B) = 0 and A is µ-null.
However, |ν|(B) = 0 implies that B is ν-null (as we proved above), so we have
ν ⊥ µ.

Problem 2. [Problem 3, page 88]

Let ν be a signed measure on (X, M).
a. L1 (ν) = L1 (|ν|).
b. If f ∈ L1 (ν), Z  Z
 
 f dν  ≤ |f | d|ν|.
 

c. If E ∈ M,  Z  
 
|ν|(E) = sup  f dν  | |f | ≤ 1
 
E

Answer :
According to the definition on page 88 of the text, the definition of L1 (ν) is
L1 (ν) = L1 (ν + ) ∩ L1 (ν − ) and if f ∈ L1 (ν), we define
Z Z Z
f dν = f dν + − f dν − .

So, consider part (a.) of the problem. If f ∈ L1 (ν), then, by definition the
integrals
Z Z
(∗) |f | dν + , |f | dν −

are finite. On the other hand, |ν| is defined by |ν|(E) = ν + (E) + ν − (E). Hence,
the equation Z Z Z
χE d|ν| = χE dν + + χE dν −

holds by definition. Thus, we can apply the usual argument to show that
Z Z Z
(∗∗) g d|ν| = g dν + + g dν −

holds for all measurable g : X → [0, ∞]: This equations holds for characteristic
functions, hence (by linearity) for simple functions and hence (by the monotone
convergence theorem) Rfor nonnegative functions. Thus, if the integrals in (∗)
are finite, the integral |f | d|ν| is finite, soRf ∈ L1 (|ν|).
Conversely, if f ∈ L1 (|ν|), the integral |f | d|ν| is finite, and so by (∗∗), the
integrals in (∗) are finite, so f ∈ L1 (ν).

2
 Next consider part (b.) of the problem. If f ∈ L1 (ν), then by definition
Z Z Z Z Z Z Z
f dν = f dν + − f dν − = f + dν + − f − dν + − f + dν − + f − dν − ,

where all of the integrals on the right are positive numbers. Hence, by the
triangle inequality,
Z  Z Z Z Z
 f dν  ≤ f + dν + + f − dν + + f + dν − + f − dν −
 
 
Z Z
= (f + + f − ) dν + + (f + + f − ) dν −
Z Z
= |f | dν + + |f | dν −
Z
= |f | d|ν|

Finally, consider part (c.) of the problem. If |f | ≤ 1, then for any E ∈ M,

|f |χE ≤ χE . Thus, we have
Z  Z
 
 f dν  ≤ |f | d|ν|
 
E E
Z
= |f |χE d|ν|
ZX
≤ χE d|ν|
X
= |ν|(E).
Thus, the sup in part (c.) is ≤ |ν|(E).
To get the reverse inequality, let X = P ∪ N be a Hahn decomposition of X
with respect to ν, and recall the description of ν + and ν − in terms of P and N
from the last problem. Then we have
Z
|ν|(E) = ν(E ∩ P ) − ν(E ∩ N ) = (χE∩P − χE∩N ) dν.

But, |χE∩P − χE∩N | ≤ 1 (P and N are disjoint). This shows that |ν|(E) is an
element of the set we’re supping over, so |ν|(E) ≤ the sup.

Problem 3. [Problem 4, page 88]

If ν is a signed measure and λ, µ are positive measures such that ν = λ − µ,
then λ ≥ ν + and µ ≥ ν − .
Answer :
Let X = P ∪ N be a Hahn decomposition of X with respect to ν. Then for any
measurable set E, we have
(∗) ν + (E) = ν(E ∩ P ) = λ(E ∩ P ) − µ(E ∩ P ).

3
If ν + (E) = ∞, this equation shows λ(E ∩ P ) = ∞, and so λ(E) = ∞. Thus,
ν + (E) ≥ λ(E) is true in this case. If ν + (E) < ∞, then both numbers on the
right of (∗) must be finite, and (∗) gives ν + (E) + µ(E ∩ P ) = λ(E ∩ P ). Thus,
we have
ν + (E) ≤ ν + (E) + µ(E ∩ P ) = λ(E ∩ P ) ≤ λ(E).
The argument for the other inequality is similar. We have

ν − (E) = −ν(E ∩ N ) = −λ(E ∩ N ) + µ(E ∩ N )

If ν − (E) = ∞, then we must have µ(E) ≥ µ(E ∩ N ) = ∞, so µ(E) ≥ ν − (E).

Otherwise, every thing must be finite and

ν − (E) ≤ ν − (E) + λ(E ∩ N ) = µ(E ∩ N ) ≤ µ(E).

Problem 4. [Problem 5, page 88] If ν1 and ν2 are signed measures that

both omitted the value +∞ or −∞, then |ν1 + ν2 | ≤ |ν1 | + |ν2 |.
(Use Exercise 4.)
Answer :
The hypothesis that ±∞ is omitted assures that the signed measure ν1 + ν2 is
defined.
We have v1 = ν1+ − ν1− and ν2 = ν2+ − ν2− . Adding these equations gives

ν1 + ν2 = (ν1+ + ν2+ ) − (ν1− + ν2− ),

where the expressions in parentheses are positive measures. Thus, by Exercise 4,

we have

(ν1 + ν2 )+ ≤ ν1+ + ν2+

(ν1 + ν2 )− ≤ ν1− + ν2−

and adding these inequalities gives |ν1 + ν2 | ≤ |ν1 | + |ν2 |.

Problem 5. [Problem 12, page 92]

For j = 1, 2, let µj , νj be σ-finite measures on (Xj , Mj ) such that νj  µj .
then ν1 × ν2  µ1 × µ2 and
d(ν1 × ν2 ) dν1 dν2
(x1 , x2 ) = (x1 ) (x2 ).
d(µ1 × µ2 ) dµ1 dµ2
Answer :
We first want to show that ν1 × ν2  µ1 × µ2 . So, suppose that E ∈ M1 ⊗ M2
and (µ1 × µ2 )(E) = 0. Recall that the x1 -section of E is defined by

Ex1 = { x2 ∈ X2 | (x1 , x2 ) ∈ E } .

4
By Theorem 2.36 on page 66 of the text, the function x1 7→ µ2 (Ex1 ) is measur-
able and Z
0 = (µ1 × µ2 )(E) = µ2 (Ex1 ) dµ1 (x1 )
X1

Thus, µ2 (Ex1 ) = 0 for µ1 -almost all x1 . In other words, there is a µ1 -null set
N ⊆ X1 so that µ2 (Ex1 ) = 0 if x1 ∈/ N . If x1 ∈ / N , we must have ν2 (Ex1 ) = 0,
since ν2  µ2 . We also have ν1 (N ) = 0, since ν1  µ1 . Thus, x1 7→ ν2 (Ex1 ) is
zero almost everywhere with respect to ν1 . Thus, we have
Z
(ν1 × ν2 )(E) = ν2 (Ex1 ) dν1 (x1 ) = 0.
X1

Thus, (µ1 × µ2 )(E) = 0 =⇒ (ν1 × ν2 )(E) = 0, so ν1 × ν2  µ1 × µ2 .

For notational simplicity, let

d(ν1 × ν2 ) dν1 dν2

f= , f1 = , f2 = .
d(µ1 × µ2 ) dµ1 dµ2

The, we have
Z
(∗) (ν1 × ν2 )(E) = f (x1 , x2 ) d(µ1 × µ2 )(x1 , x2 )
E

and f is the unique (up to equality almost everywhere) function with this prop-
erty. Of course we have
Z
ν1 (A) = f1 (x1 ) dµ1 (x1 )
ZA
ν2 (B) = f2 (x2 ) dµ2 (x2 )
B

if A ⊆ X1 and B ⊆ X2 are measurable.

On the other hand, the functions (x1 , x2 ) 7→ f1 (x1 ) and (x1 , x2 ) 7→ f2 (x2 )
are measurable, so we can define a measure λ on X1 × X2 by
Z
(∗∗) λ(E) = f1 (x1 )f2 (x2 ) d(µ1 × µ2 )(x1 , x2 ).
E

Suppose that A1 ⊆ X1 and A2 ⊆ X2 . Then, we can use Tonelli’s Theorem to

5
calculate λ(A1 × A2 ) as follows:
Z
λ(A1 × A2 ) = f1 (x1 )f2 (x2 ) d(µ1 × µ2 )(x1 , x2 )
A ×A
Z 1 2
= χA1 ×A2 (x1 , x2 )f1 (x1 )f2 (x2 ) d(µ1 × µ2 )(x1 , x2 )
ZX1 ×X2
= χA1 (x1 )χA2 (x2 )f1 (x1 )f2 (x2 ) d(µ1 × µ2 )(x1 , x2 )
ZX1 ×X
Z 2
= χA1 (x1 )χA2 (x2 )f1 (x1 )f2 (x2 ) dµ2 (x2 ) dµ1 (x1 )
ZX1 X2 Z
= χA1 (x1 )f1 (x1 ) χA2 (x2 )f2 (x2 ) dµ2 (x2 ) dµ1 (x1 )
X1 X2
Z
= χA1 (x1 )f1 (x1 )ν2 (A2 ) dµ1 (x1 )
X1
Z
= ν2 (A2 ) χA1 (x1 )f1 (x1 ) dµ1 (x1 )
X1
= ν1 (A1 )ν2 (A2 ).

Thus we have

λ(A1 × A2 ) = ν1 (A1 )ν2 (A2 ) = (ν1 × ν2 )(A1 × A2 ).

for all measurable rectangles A1 × A2 . Since our measures are σ-finite, this
shows that λ = ν1 × ν2 , see the remark at the bottom of page 64 in the
text. Comparing (∗) and (∗∗) and using the uniqueness in (∗) then shows
that f (x1 , x2 ) = f1 (x1 )f2 (x2 )

Problem 6. [Problem 16, page 92] Suppose that µ, ν are measures on

(X, M) with ν  µ and let λ = µ + ν. If f = dν/dλ, then
0 ≤ f < 1 µ-a.e. and dν/dµ = f /(1 − f ).
Answer :
It’s clear that ν  λ, so f = dν/dλ makes sense.
We want to show 0 ≤ f < 1 µ-a.e. Suppose, for a contradiction, that
the is a set E with µ(E) > 0 and f ≥ 1 on E. On the on hand, we have
ν(E) < ν(E) + µ(E) = λ(E). On the other hand, f χE ≥ χE , so
Z Z Z
ν(E) = f dλ = χE f dλ ≥ χE dλ = λ(E),
E

a contradiction.
If E is measurable, we have
Z Z Z Z
1 dν = ν(E) = f dλ = f dµ + f dν,
E E E E

6
which yields Z Z
(1 − f ) dν = f dµ,
E E
or, in other words, Z Z
χE (1 − f ) dν = χE f dµ.

By the usual procedure, we can extend this equation from characteristic func-
tions to nonnegative functions. Thus, we have
Z Z
g(1 − f ) dν = gf dµ

for all nonnegative measurable functions g. Hence, for any measurable set E,
we have
Z Z
(∗) g(1 − f ) dν = gf dµ
E E

for all nonnegative measurable functions g.

Since 0 ≤ f < 1 µ-a.e. the function 1/(1 − f ) is defined and nonnegative
µ-a.e. Since ν  µ, 1/(1 − f ) is also defined and nonnegative ν-a.e. Thus, we
can set g = 1/(1 − f ) in (∗). This gives
Z
f
ν(E) = dµ,
E 1 − f

for all measurable E, whence dν/dµ = f /(1 − f ).

Problem 7. [Problem 18, page 94] Prove Proposition 3.13c.

In other words, suppose that ν is a complex measure on (X, M). Then
L1 (ν) = L1 (|ν|) and if f ∈ L1 (ν),
Z  Z
 
 f dν  ≤ |f | d|ν|.
 

Answer :
I seem to have been very confused the day I tried to do this in class!
Let z = x + iy be a complex number, where x, y are real. We have

x2 , y 2 ≤ x2 + y 2 = |z|2

so taking square roots gives |x|, |y| ≤ |z|. One the other hand, we have

|z|2 = x2 + y 2 = |x|2 + |y|2 ≤ |x|2 + 2|x||y| + |y|2 = [|x| + |y|]2 ,

and so |z| ≤ |x| + |y|.

7
 Now lets begin the solution of the problem. By the definition on page 93 of
the text, L1 (ν) is defined to be L1 (νr ) ∩ L1 (νi ) and for f ∈ L1 (ν), we define
Z Z Z
f dν = f dνr + i f dνi .

On the other hand, we know from Exercise 3 on page 88 of the text (one of
the problems in this set) that for a signed measure λ, L1 (λ) = L1 (|λ|) and for
f ∈ L1 (λ), Z  Z
 
 f dλ ≤ |f | d|λ|.
 

Hence in our present situation, we know L1 (ν) = L1 (|νr |) ∩ L1 (|νi |).

Let’s recall the reasoning of part b. of Proposition 3.13. We can find some
finite positive measure µ such that νr and νi are absolutely continuous with
respect to µ (µ = |νr | + |νi | will do). Then, by the Lebesgue-Radon-Nikodym
Theorem, there are real-valued measurable functions f and g such that dνr =
f dµ and dνi = g dµ. Then, we have
Z
(∗) ν(E) = h dµ
E

where h = f + ig. From this equation, we have d|ν| = |h| dµ, by definition.
Since |f |, |g| ≤ |h| we have
Z Z
|νr |(E) = |f | dµ ≤ |h| dµ = |ν|(E)
ZE ZE
|νi |(E) = |g| dµ ≤ |h| dµ = |ν|(E)
E E

so we certainly have νr , νi  |ν|. Thus, once again, we can find real-valued

functions ϕ and η such that dνr = ϕ d|ν| and dνi = η d|ν|. If ϕ = ψ + iη, then
Z
ν(E) = ϕ d|ν|.
E

But then, comparing with (∗), we have

Z Z
ϕ|h| dµ = h dµ.
E E

By the uniqueness part of the Lebesgue-Radon-Nikodym Theorem, ϕ|h| = h,

µ-a.e., and hence |ν|-a.e. On the other hand if Z is the set where h = 0, then
Z Z
|ν|(Z) = |h| dµ = 0 dµ = 0
Z Z

so h 6= 0, |ν|-a.e. This shows that |ϕ| = 1, |ν|-a.e., so we can conclude that

|ψ|, |η| ≤ |ϕ| = 1, |ν|-a.e., and 1 = |ϕ| ≤ |ψ| + |η|, |ν|-a.e. Finally, we have
d|νr | = |ψ| d|ν| and d|νi | = |η| d|ν|.

8
 Now, suppose that f ∈ L1 (|ν|). Then
Z
|f | d|ν| < ∞

Since |f ||ψ| ≤ |f |, |ν|-a.e., we have

Z Z
|f ||ψ| d|ν| ≤ |f | d|ν| < ∞

but the integral on the left is Z

|f | d|νr |,

so f ∈ L1 (|νr |). Similarly, f ∈ L1 (|νi |)

Conversely, suppose that f ∈ L1 (|νr |) ∩ L1 (|νi |) Then we have
Z Z
|f ||ψ| d|ν| = |f | d|νr | < ∞
Z Z
|f ||η| d|ν| = |f | d|νi | < ∞.

and so we have Z
|f |[|ψ| + |η|] d|ν| < ∞.

However, we have 1 ≤ |ψ| + |η|, |ν|-a.e., so

Z Z
|f | d|ν| ≤ |f |[|ψ| + |η|] d|ν| < ∞.

and so f ∈ L1 (|ν|).
Finally, suppose that f ∈ L1 (ν) = L1 (|ν|), we then have
Z  Z Z 
   
 f dν  =  f dνr + i f dνi  by definition
   
Z Z 
 
=  f ψ d|ν| + i f η d|ν|
Z 
 
=  f [ψ + iη] d|ν|

Z 
 
=  f ϕ d|ν|

Z
≤ |f ||ϕ| d|ν|
Z
= |f | d|ν|,

since |ϕ| = 1, |ν|-a.e. This completes the proof.

9
Problem 8. [Problem 20, page 94] If ν is a complex measure on (X, M)
and ν(X) = |ν|(X), then ν = |ν|.
Answer :
Let f = dν/d|ν|, so we have
Z
(∗) ν(E) = f d|ν|
E

for all measurable sets E, and we know from Proposition 3.13 on page 94 of the
text that |f | = 1 |ν|-a.e. We may as well suppose that |f | = 1 everywhere.
We can write the complex function f as f = g + ih, where g and h are
real-valued. We can also write ν = νr + iνi for finite signed measures νr and νi .
Thus, we have
Z Z Z
νr (E) + iνi (E) = ν(E) = f d|ν| = g d|ν| + i h d|ν|.
E E E

Comparing real and imaginary parts, we get

Z
νr (E) = g d|ν|
ZE
νi (E) = h d|ν|.
E

By hypothesis, we have ν(X) = |ν|(X), so

Z Z
g d|ν| + i h d|ν| = |ν|(X).
X X

Since the right-hand side is real, the imaginary part of the left-hand side must
be zero, and we have
Z Z
g d|ν| = |ν|(X) = 1 d|ν|,
X X

and so
Z
(∗∗) 0= (1 − g) d|ν|.
X

Since g is the real part of f , we have

g ≤ |g| ≤ |g + ih| = |f | = 1

so 1 − g ≥ 0. But then (∗∗) shows that 1 − g = 0 a.e., so g = 1 a.e.

We then have 1 = |f |2 = g 2 + h2 = 1 + h2 a.e., so h = 0 a.e. and hence
f = g = 1 a.e. Putting this in (∗) shows that ν = |ν|.

10
Problem 9. [Problem 20, page 94] Let ν be a complex measure on (X, M).
If E ∈ M, define
 
X n [n 
(A) µ1 (E) = sup |ν(Ej )| | n ∈ N, E1 , . . . , En disjoint, E = Ej ,
 
j=1 j=1
 
X ∞ [∞ 
(B) µ2 (E) = sup |ν(Ej )| | E1 , E2 , . . . disjoint, E = Ej ,
 
j=1 j=1
 Z  
 
(C) µ3 (E) = sup  f dν  | |f | ≤ 1 .
E

Then µ1 = µ2 = µ3 = |ν|.
Answer :
Although the author didn’t explicitly say so, the sets Ej in (A) and (B) should,
of course, be measurable and the functions f in (C) should be measurable. Also
note that the book has a typo in (C) ( dµ instead of dν).
As suggested in the book, we first prove that µ1 ≤ µ2 ≤ µ3 ≤ µ1 and then
that µ3 = |ν|.
We have µ1 ≤ µ2 , because the sums in (A) are among the sums in (B): given
a finite sequence E1 , . . . , En of disjoint sets whose union is E, just set Ej = ∅
for j > n. Then E1 , E2 , . . . is an infinite sequence of disjoint sets whose union
is E and
Xn X∞
|ν(Ej )| = |ν(Ej )|.
j=1 j=1

Thus, the set of numbers in (A) is a subset of the set of numbers in (B). Since
the sup over a larger set is larger, µ1 ≤ µ2 .
To see that µ2 ≤ µ3 , let E1 , E2 , . . . be a sequence of disjoint sets whose
union is E. Now each ν(Ej ) is a complex number, so there is a complex number
ωj such that |ωj | = 1 and ωj ν(Ej ) = |ν(Ej )|. Indeed, we can say
( |ν(E )|
j
, ν(Ej ) 6= 0
ωj = ν(Ej )
1, ν(Ej ) = 0.
Consider the function f defined by
∞
X
f= ωj χEj .
j=1

If x ∈
/ E, f (x) = 0. If x ∈ E, it is in exactly one Ej and then |f (x)| = |ωj | = 1.
Thus, |f | ≤ 1 and the partial sums
n
X
fn = ωj χEj
j=1

11
satisfy |fn | ≤ 1 by the same reasoning. We then have
Z Z  Z  Z
   
 fn dν − f dν  =  (fn − f ) dν  ≤ |fn − f | d|ν| → 0
   

by the dominated convergence theorem, since fn → f pointwise and |fn − f | is

bounded above by the constant function 2, which is integrable with respect to
the finite measure |ν|. We then have
Z Z
f dν = lim fn dν
E n→∞
∞ Z
X
= ωj χEj dν
j=1 E

X∞
= ωj ν(Ej )
j=1
X∞
= |ν(Ej )|.
j=1
R
since E
f dν is positive, we have
Z  ∞
  X
 f dν  =
  |ν(Ej )|.
E j=1

Thus, the set of numbers in (B) is a subset of the set in (C), so µ2 ≤ µ3 .

Now we want to show that µ3 ≤ µ1 . First, we use Proposition 3.13 (and
Exercise 18) to show that µ3 is finite. If |f | ≤ 1, then
Z Z
|f | d|ν| ≤ 1 d|ν| = |ν|(X) < ∞,

so f ∈ L1 (|ν|) = L1 (ν) and

Z  Z Z
 
 f dν  ≤ |f | d|ν| ≤ 1 d|ν| = |ν|(E).
 
E E E

Thus, |ν|(E) is an upper bound for the set of numbers in (C), so µ3 (E) ≤
|ν|(E) ≤ |ν|(X) < ∞.
To show that µ3 ≤ µ1 , let ε > 0 be given. Then we can find some f with
|f | ≤ 1 such that Z 
 
µ3 (E) − ε <  f dν ,
E
which implies
Z 
 
(∗) µ3 (E) ≤  f dν  + ε.

E

12
 We approximate f by a simple function as follows. Let

D = { z ∈ C | |z| ≤ 1 }

be the closed unit disk in the complex plane. Since |f | ≤ 1, all the values of f
are in D. The collection of balls

{ B(ε, z) | z ∈ D }

is an open cover of D. Since D is compact, there is a finite subcover, say

{ B(ε, zj ) | j = 1, . . . , m } .

Define Bj = f −1 (B(ε, zj )) ⊆ X, which is measurable since f is measurable. The

union of the sets Bj is X. The sets Bj won’t be disjoint, but we can use the
usual trick and define

A1 = B1 ,
j−1
[
Aj = Bj \ Bk
k=1
S
to get a disjoint collection of sets with Aj ⊆ Bj and j Aj = X. Define a simple
function ϕ by
m
X
ϕ= zj χAj .
j=1

This function takes on the values zj , which are in D, so |ϕ| ≤ 1. If x ∈ X it is

in exactly one of the sets Aj . But then f (x) ∈ B(ε, zj ), so

|f (x) − ϕ(x)| = |f (x) − zj | < ε.

Thus, |f − ϕ| < ε. We then have

Z  Z  Z  Z 
       
 f dν  −  ϕ dν  ≤  f dν  −  ϕ dν 
       
E E
Z E Z
E

 
≤ f dν − ϕ dν 
ZE E

≤ |f − ϕ| d|ν|
ZE
≤ ε d|ν|
E
= ε|ν|(E).

Thus, we have Z  Z 
   
 f dν  ≤  ϕ dν  + ε|ν|(E).
   
E E

13
Substituting this is (∗) gives
Z 
 
(∗∗) µ3 (E) ≤  ϕ dν  + ε + ε|ν|(E).
E

Now define Ej = Aj ∩ E, j = 1, . . . , m. Then the Ej ’s are a finite sequence

of disjoint sets whose union is E and we have
Z  Z X m  
   
 ϕ dν  =  z j Aj dν 
χ 
  
E E j=1

m
X Z 

=
 zj χAj dν 
j=1 E
m Z 
X 
=
 zj χE χAj dν 
j=1
m Z 
X 
=  zj χEj dν 
j=1
m 
X 
=
 zj ν(Ej )
j=1
m
X
≤ |zj ||ν(Ej )|
j=1
Xm
≤ |ν(Ej )|
j=1

≤ µ1 (E).

Substituting this into (∗∗) gives

µ3 (E) ≤ µ1 (E) + [1 + |ν|(E)]ε.

Since ε > 0 was arbitrary, we conclude that µ3 (E) ≤ µ1 (E).

We’ve now shown that µ1 = µ2 = µ3 . The last step is to show that µ3 = |ν|.
We’ve already shown above that µ3 (E) ≤ |ν|(E). To get the reverse inequality,
let g = dν/d|ν|. We know that |g| = 1 |ν|-a.e., and we may as well assume
that |g| = 1 everywhere (by modifying it on a set of measure zero). Then the
conjugate ḡ of g satisfies |ḡ| = 1 ≤ 1 and so is one of the functions in the
definition of µ3 , so Z 
 
 ḡ dν  ≤ µ3 (E).
 
E

14
But, from the definition of g,
Z  Z 
   
 ḡ dν  =  ḡg d|ν|
   
E
 ZE 
2
 
= |g| d|ν|

 ZE 
 
=
 1 d|ν|
E
= |ν|(E).

Thus, |ν|(E) ≤ µ3 (E), so µ3 = |ν|, and we’re done.

15