302 CHAPTER 10. THE ABSTRACT LEBESGUE INTEGRAL

Thus k→ f k (ω) is increasing. Let f (ω) ≡ limk→∞ f k (ω). Thus f = 0 on S. Now letE be measurable. Thus µ (E) = µ (E ∩S)+ µ

(E ∩SC

), similar for λ and λ

(E ∩SC

)=

∑n λ(E ∩SC ∩∆k

n). To save space, let Ẽ ≡ E ∩SC. Then using 10.16∫

XẼ f kdµ ≤∞

∑n=0

αkn+1µ

(Ẽ ∩∆

kn

)≤

∑n=0

αknµ

(Ẽ ∩∆

kn

)+

∑n=0

2−kµ

(Ẽ ∩∆

kn

)≤

∑n=0

ν

(Ẽ ∩∆

kn

)+2−k

µ(Ẽ)= ν

(Ẽ)+2−k

µ(Ẽ)≤∫

XẼ f kdµ +2−kµ(Ẽ)

(10.18)

From the monotone convergence theorem it follows ν(Ẽ)=∫

XẼ f dµ =∫

XE f dµ .This proves most of the following theorem which is the Radon Nikodym theorem.

Theorem 10.13.7 Let ν and µ be finite measures defined on a measurable space(Ω,F ). Then there exists a measurable set S with µ (S) = 0 and a nonnegative measurablefunction ω → f (ω) such that ν (E) =

∫E f dµ +ν (E ∩S) .

Proof: Let S be defined in Lemma 10.13.6 so S≡Ω\(∪nNk

n)

and µ (S) = 0. If E ∈F ,and f as described above,

ν (E) = ν(E ∩SC)+ν (E ∩S) =

∫E∩SC

f dµ +ν (E ∩S) =∫

Ef dµ +ν (E ∩S)

Thus if E ⊆ SC, we have ν (E) =∫

E f dµ . ■

Definition 10.13.8 Let µ,ν be finite measures on (Ω,F ). Then ν≪ µ means thatwhenever µ (E) = 0, it follows that ν (E) = 0.

Sometimes people write f = dλ

dµ, in the case ν ≪ µ and this is called the Radon

Nikodym derivative.

Proposition 10.13.9 If ν ,µ are finite measures and ν ≪ µ, then there exists nonneg-ative measurable f such that ν (E) =

∫E f dµ.

Proof: In Theorem 10.13.7, ν (E ∩S) = 0 because µ (S) = 0 and ν ≪ µ . ■

Definition 10.13.10 Let S be in the above theorem. Then

ν || (E)≡ ν(E ∩SC)= ∫

E∩SCf dµ =

∫E

f dµ

while ν⊥ (E)≡ ν (E ∩S) . Thus ν ||≪ µ and ν⊥ is nonzero only on sets which are containedin S which has µ measure 0.

Corollary 10.13.11 In the above situation, let λ be a signed measure and let λ ≪ µ

meaning that if µ (E) = 0⇒ λ (E) = 0. Here assume that µ is a finite measure. Then thereexists h ∈ L1 such that λ (E) =

∫E hdµ .

Proof: Let P∪N be a Hahn decomposition of λ . Let

λ+ (E)≡ λ (E ∩P) , λ− (E)≡−λ (E ∩N) .

Then both λ+ and λ− are absolutely continuous measures and so there are nonnegative h+and h− with λ− (E) =

∫E h−dµ and a similar equation for λ+. Then 0 ≤ −λ (Ω∩N) ≤

λ− (Ω) < ∞, similar for λ+ so both of these measures are necessarily finite. Henceboth h− and h+ are in L1 so h ≡ h+− h− is also in L1 and λ (E) = λ+ (E)− λ− (E) =∫

E (h+−h−)dµ . ■

302 CHAPTER 10. THE ABSTRACT LEBESGUE INTEGRALThus k - f*(q) is increasing. Let f(@) = limy;..f*(@). Thus f =0 on S. Now letE be measurable. Thus p (E) = w(ENS)+u (ENS), similar for A and A (ENS°) =yA (E nse M Ak) . To save space, let E = EM S©. Then using 10.16 %efiau < y ok (Enat) < y ok (Enas) + y 2 *u (Enas)n=0 n=0 n=0<Yv (nal) +2-*u (2) =v(E)+2*u (E) < (| %efhan +2 (Z) (10.18)n=0From the monotone convergence theorem it follows v (E) = f 2¢fdu =f 2efdu.This proves most of the following theorem which is the Radon Nikodym theorem.Theorem 10.13.7 Let v and u be finite measures defined on a measurable spaceLl(Q, .F). Then there exists a measurable set S with ut (S) =0 and a nonnegative measurablefunction @ > f (@) such that v(E) = fy, fdu+v (ENS).Proof: Let S be defined in Lemma 10.13.6 so S=Q\ (U,N*) and pw (S) =0. If E € F,and f as described above,V(E)=v(ENS*)+v(ENS)= | fau+v(Ens)= ff fdu+v(ENs)Thus if E C S©, we have v(E) = J, fdu.Definition 10.13.8 Le: LL, v be finite measures on (Q, F). Then Vv < [means thatwhenever U (E) = 0, it follows that v(E) = 0.daSometimes people write f = du?Nikodym derivative.in the case v < w and this is called the RadonProposition 10.13.9 If v,u are finite measures and v « UL, then there exists nonneg-ative measurable f such that v(E) = J, fdw.Proof: In Theorem 10.13.7, v(EMS) = 0 because (S) =O andv <p. HfDefinition 10.13.10 Let 8 be in the above theorem. ThenVv) (E) =v (ENS*) =| fdu= | fanwhile v, (E) =V(ENS). Thus v\| < and v, is nonzero only on sets which are containedin S which has & measure 0.Corollary 10.13.11 In the above situation, let A be a signed measure and let’ <meaning that if U(E) =0 => A (E) =0. Here assume that p is a finite measure. Then thereexists h € L' such that A (E) = J, hd.Proof: Let PUN be a Hahn decomposition of A. LetA+(E)=A(ENP), A_(E)=-A(ENN).Then both A, and A_ are absolutely continuous measures and so there are nonnegative h+and h_ with A_ (E) = f,h—du and a similar equation for A. Then 0 < —A (QNN) <A_(Q) < c, similar for A+ so both of these measures are necessarily finite. Henceboth A_ and hy are in L' soh=h, —h_ is also in L! and A(E) =A, (E)—A_(E) =e(ty — hy.