9.11. RADON NIKODYM THEOREM 219

Clearly a Hahn decomposition is not unique. For example, you could have obtaineda different Hahn decomposition if you had considered disjoint negative sets F for whichλ (F)< 0 in the above argument. I will only use the case where ν≪ µ which is to say thatν is absolutely continuous with respect to µ which is defined next.

Definition 9.11.6 Let µ,ν be finite measures on (Ω,F ). Then ν ≪ µ means thatwhenever µ (E) = 0, it follows that ν (E) = 0.

Let k ∈ N,{

αkn}∞

n=0 be equally spaced points αkn = 2−kn. Then αk

2n = 2−k (2n) =2−(k−1)n≡ αk−1

n and αk+12n ≡ 2−(k+1)2n = αk

n. Similarly Nk+12n = Nk

n because these dependon the αk

n. Also let(Pk

n ,Nkn)

be a Hahn decomposition for the signed measure ν −αknµ

where ν ,µ are two finite measures. Now from the definition, Nkn+1 \Nk

n = Nkn+1∩Pk

n . Also,Nn ⊆ Nn+1 for each n and we can take N0 = /0 because ν (N0) ≤ 0 Then

{Nk

n+1 \Nkn}∞

n=0covers all of Ω except for a set of ν measure 0.

Lemma 9.11.7 Let S≡Ω\(∪nNk

n)= Ω\

(∪nNl

n)

for any l. Then µ (S) = 0.

Proof: S = ∩nPkn so for all n,ν (S)−αk

nµ (S) ≥ 0. But letting n→ ∞, it must be thatµ (S) = 0. ■

By the assumption that ν ≪ µ, we can neglect S because this also implies ν (S) = 0.Thus, asside from a set of µ and ν measure zero, Ω = ∪nNk

n .As just noted, if E ⊆ Nk

n+1 \Nkn , then

ν (E)−αknµ (E)≥ 0≥ ν (E)−α

kn+1µ (E) , so α

kn+1µ (E)≥ ν (E)≥ α

knµ (E) (9.12)

Nkn

Nkn+1

αkn+1µ(E)≥ ν(E)≥ αk

nµ(E)

Then define f k (ω)≡ ∑∞n=0 αk

nX∆kn(ω) where ∆k

m ≡ Nkm+1 \Nk

m. Thus,

f k =∞

∑n=0

αk+12n X(Nk+1

2n+2\Nk+12n ) =

∑n=0

αk+12n X

∆k+12n+1

+∞

∑n=0

αk+12n X

∆k+12n

≤∞

∑n=0

αk+12n+1X∆

k+12n+1

+∞

∑n=0

αk+12n X

∆k+12n

= f k+1 (9.13)

Thus k → f k (ω) is increasing. Let f (ω) ≡ limk→∞ f k (ω). Then using 9.12, if E is ameasurable set,∫

XE f kdµ ≤∞

∑n=0

αkn+1µ

(E ∩∆

kn

)≤

∑n=0

αknµ

(E ∩∆

kn

)+

∑n=0

2−kµ

(E ∩∆

kn

)

≤∞

∑n=0

ν

(E ∩∆

kn

)+2−k

µ (E) = ν (E)+2−kµ (E)≤

∫XE f kdµ +2−k

µ (E) (9.14)

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

XE f dµ .This proves the following major theorem called the Radon Nikodym theorem.

9.11. RADON NIKODYM THEOREM 219Clearly a Hahn decomposition is not unique. For example, you could have obtaineda different Hahn decomposition if you had considered disjoint negative sets F for whichA (F) <0 in the above argument. I will only use the case where v < which is to say thatv is absolutely continuous with respect to which is defined next.Definition 9.11.6 Le Ll, v be finite measures on (Q,-#). Then V < [ means thatwhenever LL (E) = 0, it follows that v (E) = 0.Let k EN, {ak} 9 be equally spaced points af =2~'n. Then aS, = 2-*(2n) =2-"—Dn = ok! and aft! =2-()2n = ak, Similarly N¥*! = Nk because these dependon the a. Also let (Pk, N*) be a Hahn decomposition for the signed measure v—okuwhere V, LU are two finite measures. Now from the definition, N7 ian \ Nk = Ni rl PK. Also,Nn © Nn+1 for each n and we can take No = 0 because v (No) < 0 Then {Nk NEScovers all of Q except for a set of v measure 0.n+1Lemma 9.11.7 Let S = Q\ (UnNk) = Q\ (U,Ni) for any 1. Then p(S) =0.Proof: S = ,,P* so for all n,v(S) — aku (S) > 0. But letting n > o>, it must be thatL(S) =0.By the assumption that v < 1, we can neglect S because this also implies v(S) = 0.Thus, asside from a set of and v measure zero, Q = UnNE.As just noted, if E C Nk, , \ Nk, thenv(E)—oku(E) >0>v(E)—ak,,u(E), so ak, u(E) > v(E) > aku(E) (9.12)Ons M(E) = V(E) = aM (E)Then define f* (@) =”, ak X nx (@) where AKNX, \Nj,. Thus,f* = y okt! Zvks, \wk+1) -¥ atk! Be aNGan + x atk! Qe aktln=0<y Oo Zk, + x Oy Ze =f (9.13)n=0Thus k + f*(@) is increasing. Let f(@) = limy,.f*(@). Then using 9.12, if E is ameasurable set,/ Re fidy < Yok, ya (Enak) < ¥ ok (Enak) + v2" (Enak)n=0 n=0n=0<P v(Enat) +2-w(e) =v V(E)+2*u(E )< | %eftau+2 kKu(E) (9.14)n=0From the monotone convergence theorem it follows v(E) = f Ze fdu.This proves the following major theorem called the Radon Nikodym theorem.