Measure theory class notes - 13 October 2010, class 20 1

Proof of Radon-Nikodym theorem

Theorem (Radon-Nikodym). Suppose Ω is a nonempty set and A a σ-field on it. Suppose µ and
ν are σ-finite measures on (Ω, A ) such that for all A ∈ A , µ(A) = 0 implies ν(A) = 0. Then

• There exists z : Ω → [0, ∞) measurable such that for all A ∈ A , ν(A) =

R
A
z dµ.

• Such a z is unique upto a.e. equality (w.r.t. µ).

• z is integrable w.r.t. µ if and only if ν is a finite measure.

Proof. Assuming we have obtained

R the z, it is easy to see that z is integrable if and only if ν is a
finite measure, since ν(Ω) = z dµ.

Uniqueness
We
R show uniqueness R [0, ∞) such that for all A ∈ A ,
R first. Suppose there are two Rz1 , z2 : Ω →
z dµ = ν(A) = A z2 dµ. We would like to say A z1 dµ − A z2 dµ = 0, but both the integrals
A 1
might be ∞. Both µ and ν are σ-finite. Let C be a countable partition of Ω such that for all
C ∈ C, µ(C) < ∞ and ν(C) < ∞ (to obtain this we can take such a partition for µ, and such a
partition for ν, and take pairwise intersections). Fix a C ∈ C. Let

B = C ∩ {ω : z1 (ω) − z2 (ω) > 0}

Note that 1B (z1 −z2 ) = 1C (z1 −z2 )+ (evaluate both sides at ω: for ω ∈
/ C, both are 0; for ω ∈ C \B,
LHS
R is 0 andR since (z1 − z2 )(ω) ≤ 0, RHS is 0; for ω ∈ B both side are z1 (ω)
R − z2 (ω)). R ν(B) =
1B z1 dµ = 1B z2 dµ. Since ν(B) ≤ ν(C) < ∞, it is meaningful to consider 1B z1 dµ− 1B z2 dµ.
Z Z
0 = 1B z1 dµ − 1B z2 dµ
Z
= 1B (z1 − z2 ) dµ
Z
= 1C (z1 − z2 )+ dµ

If the integral of a nonnegative function is 0, it must be 0 almost everywhere. Applying to the

above:
µ({ω : ω ∈ C and z1 (ω) − z2 (ω) > 0}) = 0
We can interchange the roles of z1 and z2 in the above argument. Combining the two, we get

µ({ω : ω ∈ C and z1 (ω) − z2 (ω) 6= 0}) = 0

This holds for all C ∈ C. Since C is countable, and a countable union of null sets is a null set,

µ({ω : z1 (ω) 6= z2 (ω)}) = 0

This proves that the z promised in the theorem (if it exists) is unique upto a.e. (w.r.t. µ) equality.
Measure theory class notes - 13 October 2010, class 20 2

Existence
We now show the existence of such a z. We reduce the general case to the case when both µ and
ν are finite, and then show existence for the finite case.

Reduction to the finite measure case

Suppose we know the Radon-Nikodym theorem holds for the case when the measures involved are
finite. In general, assume µ and ν are σ-finite, and let C be a countable partition of Ω such that
for all C ∈ C, µ(C) and ν(C) are finite. Define finite measures µC and νC for all C ∈ C,

µC (A) = µ(C ∩ A), νC (A) = ν(C ∩ A)

If µC (A) = 0, then νC (A) = 0. Using the finite-measure case of the theorem, let zC : Ω → [0, ∞)
be such that for all A ∈ A and C ∈ C,
Z
νC (A) = zC dµC
A

zC and zC 1C differ at most on Ω \ C, which has 0 µC measure. So we can replace zC by zC 1C ,

and so assume that for ω ∈ / C, zC (ω) = 0. Define z : Ω → [0, ∞), by requiring that z|C = (zC )|C
P
(that is, for ω ∈ C, z(ω) = zC (ω)). z can also be described as C∈C zC - this shows that z is
measurable.
R R
If g : Ω → [0, ∞) is 0 outside C, then g dµ = g dµC . This can be proven as usual by proving
it for simple functions and then for all nonnegative measurable functions. This fact is applied to
zC in what follows. For any A ∈ A ,
X
ν(A) = ν(A ∩ C)
C∈C
X
= νC (A)
C∈C
XZ
= zC dµC
C∈C A
XZ
= zC dµ
C∈C A
Z X
= zC dµ (by taking partial sums; MCT)
A C∈C
Z
= z dµ
A

It remains to prove the theorem for the finite measure case:

Finite measures - getting hold of the z:

Now assume µ and ν are finite measures. Let

C = f : (f : Ω → [0, ∞]R̄ ), ∀A ∈ A ( A f dµ ≤ ν(A))

 R
Measure theory class notes - 13 October 2010, class 20 3

We want to identify f such that equality holds, that is, for all A ∈ A ,
R
A
f dµ = ν(A).
C is nonempty, because the zero function belongs to C. Intuitively, we want to choose the “largest”
function in C as a candidate for our z. Does such a “largest” function exist? We observe some
properties of C first:

• If f, g ∈ C, then f ∨ g (which is the pointwise maximum of f and g) belongs to C. Take any

A ∈ A . We apply the condition defining C to the sets A1 := A ∩ {ω : f (ω) ≥ g(ω)} and
A2 := A ∩ {ω : f (ω) < g(ω)}:
Z Z
f ≤ ν(A1 ), g ≤ ν(A2 )
A1 A2

Note that f 1A1 = (f ∨ g)1A1 and g1A2 = (f ∨ g)1A2 . Substituting in the above equations and
adding them, we get Z
(f ∨ g) ≤ ν(A)
A

• Suppose {fn }∞
n=1 is an increasing sequence in C. Let f be their supremum (equivalently
limit). Then f ∈ C: take any A ∈ A . {fn 1A }∞
n=1 increases to f 1A , and so by the monotone
R
R theorem the corresponding limit with integrals holds. Since each fn 1A ≤ ν(A),
convergence
we have f 1A ≤ ν(A).

We identify the “largest” function in C by considering the function with the largest integral. For
R
all f ∈ C, f dµ ≤ ν(Ω) < ∞. Let
nR o
α = sup f dµ : f ∈ C
∞
0 ≤ α ≤ ν(Ω). In particular, α 6= ∞. Let {fn }∞
R
n=1 be a sequence from C such that fn dµ n=1
converges to α (this can be done because of the supremum definition of α). Consider the partial
maxima of {fn }∞ ∞
n=1 , that is, define {gn }n=1 by

gn = max{f1 , f2 , . . . , fn }
R R R
We know that gn ∈ C, so gn dµ ≤ α. Also, fn ≤ gn , so fn dµ ≤ gn dµ ≤ α. Thus
limn→∞ gn dµ = α. R{gn }∞
R
n=1 is an increasing sequence; let z = limn→∞ gn . By the monotone
convergence theorem, z dµ = α. Since z is integrable, z is finite almost everywhere. We modify
z on a set of measure 0 suitably so that z never takes the value ∞. This z is our “largest” function
in C, as we show now:
Let g ∈ C. We will show that g ≤ z a.e. (w.r.t µ). Let A = {ω : g(ω) > z(ω)}. Let z ∨ g = h.
R R R
h ∈ C. A = {ω : h(ω) > z(ω)} = {ω : h(ω) − z(ω) > 0}. h ≥ z; h dµ ≥ z dµ = α, so h = α.
h − z is a nonnegative function whose µ integral is 0, so h − z is 0 almost everywhere. µ(A) = 0.
g ≤ z almost everywhere (w.r.t µ).
Measure theory class notes - 13 October 2010, class 20 4

Finite measures - showing that the z works

Define λ : A → [0, ∞), Z

λ(A) = ν(A) − z dµ
A
Note that this is well-defined, that is, the expression on the RHS indeed belongs to [0, ∞). λ is a
measure: clearly λ(∅) = 0. Let C ⊆ A be a countable collection of disjoint sets.
X XZ [  Z
ν(C) + z dµ = ν C + S z dµ < ∞
C∈C C∈C C C

Since the relevant series is absolutely convergent, we can rearrange terms, and
[  [  Z X XZ X Z  X
λ C =ν C − S z dµ = ν(C) − z dµ = ν(C) − z dµ = λ(C)
C C∈C C∈C C C∈C C C∈C

λ is a measure. We want to show that λ is the zero measure, and then we will be done.
If for all A ∈ A and all k ∈ N, λ(A) ≤ k1 µ(A) then since µ(A) is finite, λ(A) = 0, and we are
done. So there exists A ∈ A and k ∈ N such that λ(A) − k1 µ(A) > 0. Fix such an A and k. We
will derive a contradiction.
Call a set Z ∈ A to be good if the following two conditions hold:

• λ(Z) − k1 µ(Z) > 0.

• For all B ⊆ Z with B ∈ A , λ(B) − k1 µ(B) ≥ 0.

The first condition implies that a good set Z must have µ(Z) > 0: otherwise if µ(Z) = 0, then by
the hypothesis, ν(Z) = 0, and so λ(Z) = 0. A satisfies the first condition, but need not satisfy the
second condition.
We will identify a good set - A itself may not be a good set, but we can “scrape off” some parts
of it to get a good set. Once we identify a good set Z, we will use the function z + k1 1Z to get a
contradiction.
Let A0 = A. Let β0 = inf{λ(B) − k1 µ(B) : B ⊆ A, B ∈ A }. If β0 ≥ R 0, then A itself is a good set.
So assume β0 < 0. β0 cannot be −∞, since it is at least −(ν(Ω) + z dµ + k1 µ(Ω)). Pick B0 ⊆ A0
such that
1 1
λ(B0 ) − µ(B0 ) < β0
k 2
Put A1 = A \ B0 .

λ(A1 ) − k1 µ(A1 ) = λ(A) − λ(B0 ) − k1 µ(A) − k1 µ(B0 ) ≥ λ(A) − k1 µ(A) > 0

(because λ(B0 ) − k1 µ(B0 ) < 12 β0 < 0.)

An important observation is this: if B ⊆ A1 , then λ(B) − k1 µ(B) ≥ 21 β0 . Otherwise, we could take
a B which violates this, and B ∪ B0 would violate the infimum definition of β0 .
If A1 is a good set, we have found a good set. Otherwise, repeat with A1 what we did with A0 .
The above observation says that the β1 so obtained will satisfy β1 ≥ 12 β0 . Inductively, we construct
a sequences of sets {An }∞ ∞ ∞
n=1 and {Bn }n=1 , and a sequence of real numbers {βn }n=1 , such that the
following holds: some An is good, or for all n ∈ N,
Measure theory class notes - 13 October 2010, class 20 5

• Bn ⊆ An .

• An+1 = An \ Bn .

• βn = inf λ(B) − k1 µ(B) : B ⊆ An , B ∈ A , βn < 0.



• λ(An+1 ) − k1 µ(An+1 ) ≥ λ(An ) − k1 µ(An ).

• βn+1 ≥ 21 βn .

This immediately implies that for all n, βn ≥ 21n β0 and λ(An ) − k1 µ(An ) ≥ λ(A) − k1 µ(A) > 0.
T
If no An was good, define A∞ = n∈N An . Since λ and µ are finite measures,

λ(A∞ ) = lim λ(An ) and µ(A∞ ) = lim µ(An )

n→∞ n→∞

So λ(A∞ ) − k1 µ(A∞ ) ≥ λ(A) − k1 µ(A) > 0. If B ⊆ A∞ and B ∈ A , then since B ⊆ An ,

λ(B) − k1 µ(B) ≥ 21n β0 . Since this holds for all n, λ(B) − k1 µ(B) ≥ 0. A∞ is a good set! If some An
was good, let A∞ be that good set.
We will show that z + k1 1A∞ ∈ C. For any S ∈ A , if S ⊆ Ω \ A∞ , then
Z   Z
1
z + 1A∞ dµ = z dµ ≤ ν(S)
S k S

If S ⊆ A∞ , then

z + k1 1A∞ dµ = S z dµ + k1 µ(S)
R
R
S
R
≤ S z dµ + λ(S) (since A∞ is a good set)
= ν(S) (by definition of λ)

For general S, we just combine the part of S in A and the part in Ω \ A:

Z   Z   Z  
1 1 1
z + 1A∞ dµ = z + 1A∞ dµ+ z + 1A∞ dµ ≤ ν(S ∩A)+ν(S \A) = ν(S)
S k S∩A k S\A k

We have shown that z + k1 1A∞ ∈ C. Since A∞ is a good set, µ(A∞ ) > 0. It is not the case that
z + k1 1A∞ ≤ z a.e. (w.r.t. µ), which contradicts what we have shown earlier.
λ must be the zero measure, and so for all D ∈ A ,
Z
ν(D) = z dµ
D