11.13. RADON NIKODYM THEOREM, RADON MEASURES 343

11.13 Radon Nikodym Theorem, Radon MeasuresThe Radon Nikodym theorem is an abstract result but this will be a special version. It willgive a pointwise description in terms of the symmetric derivative of the Radon Nikodymderivative presented earlier.

Definition 11.13.1 Let λ ,µ be two Radon measures defined on F , a σ algebraof subsets of an open set U. Then λ ≪ µ means that whenever µ (E) = 0, it follows thatλ (E) = 0.

Next is a representation theorem for λ in terms of an integral involving Dµ λ .

Theorem 11.13.2 Let λ and µ be Radon measures defined on Fλ ,Fµ respec-tively, σ algebras of the open set U, then there exists a set of µ measure zero N such thatDµ λ (x) exists off N and if E ⊆ NC,E ∈ Fλ ∩Fµ , then λ (E) =

∫U(Dµ λ

)XEdµ . If

λ ≪ µ on Fλ ∩Fµ , then λ (E) =∫

E Dµ λdµ . In any case, λ (E)≥∫

E Dµ λdµ so Dµ λ isin L1

loc (Rp,µ) because λ (B)< ∞ for any ball B.

Proof: The proof is based on Lemma 11.12.2. Let E ⊆ NC where N has µ measure 0and includes the set Z along with the set where the symmetric derivative does not exist. Itcan be assumed that N is a Gδ set. Define

ln (x)≡∞

∑k=1

ank−1X(Dµ λ)

−1(In

k )(x) , un (x)≡

∑k=1

ankX(Dµ λ)

−1(In

k )(x)

where Ink ≡ ((k−1)2−n,k2−n] ≡ (an

k−1,ank ] for k,n ∈ N. Thus un (x) ≥ Dµ λ (x) > ln (x)

and un (x)− ln (x) = 2−n. Also, ln (x) increases to Dµ λ (x). Letting

Enk ≡

[x ∈ E : Dµ λ (x) ∈ In

k],

and assuming µ (E)< ∞,∫

E Dµ λdµ ∈ [∫

E lndµ,∫

E undµ]

=

[∞

∑k=1

ank−1µ (En

k ) ,∞

∑k=1

ank µ (En

k )

]⊆[∫

Elndµ,

∫E

lndµ +2−nµ (E)

](11.19)

From Lemma 11.12.2, µ(En

k

)an

k ≥ λ(En

k

)≥ an

k−1µ(En

k

)and so the interval in 11.19 con-

tains ∑∞k=1 λ

(En

k

). This equals λ (E) because of Lemma 11.12.2 which implies

λ(E ∩

{x ∈ NC : Dµ λ (x) = 0

})≤ aµ

(E ∩

{x ∈ NC : Dµ λ (x) = 0

})≤ aµ (E) , µ (E)< ∞

and since this is true for every positive a, it follows that

λ(E ∩

{x ∈ NC : Dµ λ (x) = 0

})= 0

so the sum ∑∞k=1 λ

(En

k

)= λ (E) . Then, from the monotone convergence theorem in 11.19,

one can pass to a limit and find that∫

E Dµ λdµ = λ (E) .Now if E is an arbitrary set in NC, maybe not bounded, the above shows

λ (E ∩B(0,n)) =∫

E∩B(0,n)Dµ λdµ

11.13. RADON NIKODYM THEOREM, RADON MEASURES 34311.13. Radon Nikodym Theorem, Radon MeasuresThe Radon Nikodym theorem is an abstract result but this will be a special version. It willgive a pointwise description in terms of the symmetric derivative of the Radon Nikodymderivative presented earlier.Definition 11.13.1 Le A, be two Radon measures defined on #, a © algebraof subsets of an open set U. Then A < LW means that whenever 1 (E) = 0, it follows thatA (E) =0.Next is a representation theorem for A in terms of an integral involving Dy.Theorem 11.13.2 Ler 2 and LL be Radon measures defined on F;,Fy respec-tively, 0 algebras of the open set U, then there exists a set of & measure zero N such thatDud (x) exists off N and if ECN°,E € F¥,0Fy, then A(E) = fy (Dud) Zedu. IfA <pon F270 Fy, then A(E) = fe DyAdu. In any case, A(E) > fp DyAdu so DyA isin L} . (R?, W) because 2 (B) < for any ball B.locProof: The proof is based on Lemma 11.12.2. Let E C NC where N has Lt measure 0)and includes the set Z along with the set where the symmetric derivative does not exist. Itcan be assumed that N is a Gg set. Defineol ee CCwhere I? = ((k—1)2~",k2~"] = (aj_,, aj] for k,n EN. Thus uy (x) > Dy A (@) > In (x)and un (ae )=In(@) = 27 "Also, 1, (a) increases to Dy A (x). LettingEi = |w EE: DyA(a) EX],and assuming Ll (E) < ©, fp Duddu € [Je lndu, Jp Und ul]Yat mie). Y atu (ep C tna [nate +2" (6) (11.19)From Lemma 11.12.2, pt (Ej) a? > A (E?) > a?_,u (Ej) and so the interval in 11.19 con-tains Ye_, A (Ef) . This equals A (E) because of Lemma 1 1.12.2 which implies(En{a ENC: DyA(a) =0})(E), U(E) <<and since this is true for every positive a, it follows thatA(EN{@eEN®: DyA(x)=0}) < au< ayA(EN{a ENS: DyA (x) =0}) =so the sum Y_, A (E?) =A (E). Then, from the monotone convergence theorem in 11.19,one can pass to a limit and find that [; DuAdu =A (E).Now if £ is an arbitrary set in NC, maybe not bounded, the above showsA (ENB(O,n)) = ono Duka