272 CHAPTER 9. MEASURES AND MEASURABLE FUNCTIONS

This defines ψ2 (ω) on Ωn and so it defines ψ2 on Ω satisfying 9.25. Continue this way,obtaining ψk a measurable function such that

dist(ψk (ω) ,Γ(ω))<1

2k−1 , d(ψk+1 (ω) ,ψk (ω)

)<

12k−2 .

Then for each ω,{ψk (ω)} is a Cauchy sequence of measurable functions converging toa point, σ (ω) ∈ Γ(ω). This has shown that if Γ is measurable, there exists a measurableselection, σ (ω) ∈ Γ(ω). Of course, if Γ(ω) is closed, then σ (ω) ∈ Γ(ω). Note that thishad nothing to do with any measure.

It remains to show that there exists a sequence of these measurable selections σn suchthat the conclusion of 2.) holds. To do this define for a single ω ∈Ω

Γni (ω)≡{

Γ(ω)∩B(xn,2−i

)if Γ(ω)∩B

(xn,2−i

)̸= /0

Γ(ω) otherwise when there is empty intersection .

The following picture illustrates Γni(ω) when ω is such that there is nonempty intersection.Also, given x ∈ Γ(ω), and i, there is xn from the countable dense set such that the situationof the picture occurs.

•xn

•x ∈ Γ(ω)Γ(ω)

Γni(ω)

B(xn,2−i)

Is Γni measurable? If so, then from the above, it has a measurable selection σni andthe set of these σni must have the property that {σni (ω)}n,i is dense in Γ(ω) for each ω .

Let U be open. Then

{ω : Γni (ω)∩U ̸= /0}={

ω : Γ(ω)∩B(xn,2−i)∩U ̸= /0

}∪[{

ω : Γ(ω)∩B(xn,2−i)= /0

}∩{ω : Γ(ω)∩U ̸= /0}

]={

ω : Γ(ω)∩B(xn,2−i)∩U ̸= /0

}∪[(

Ω\{

ω : Γ(ω)∩B(xn,2−i) ̸= /0

})∩{ω : Γ(ω)∩U ̸= /0}

],

a measurable set. Thus Γni is measurable as hoped.By what was just shown, there exists σni, a measurable function such that σni (ω) ∈

Γni (ω) ⊆ Γ(ω) for all ω ∈ Ω. If x ∈ Γ(ω), then x ∈ B(

xn,2−(i+1))

whenever xn is close

enough to x. Thus both x,σn(i+1) (ω) are in B(xn,2−(i+1)

)and so

∣∣σn(i+1) (ω)− x∣∣< 2−i.

It follows that condition 2.) holds with the countable dense subset of Γ(ω) being the{σni (ω)}. Note that this had nothing to do with a measure.

Now consider why 2.)⇒ 1.). We have {σn (ω)} ⊆ Γ(ω) and σn is measurable and∪nσn (ω) equals Γ(ω). Why is Γ a measurable multifunction? Let U be an open set

Γ− (U) ≡ {ω : Γ(ω)∩U ̸= /0}=

{ω : Γ(ω)∩U ̸= /0

}= ∪nσ

−1n (U) ∈F ■

272 CHAPTER 9. MEASURES AND MEASURABLE FUNCTIONSThis defines y, (@) on Q, and so it defines yy on Q satisfying 9.25. Continue this way,obtaining y, a measurable function such thatdist (yw, (@) ,T'(@)) < a d (Wix1(@), We (@)) < eTThen for each @, {wW,(@)} is a Cauchy sequence of measurable functions converging toa point, o(@) € I'(@). This has shown that if I is measurable, there exists a measurableselection, o(@) € I (@). Of course, if [(@) is closed, then o(@) € T'(@). Note that thishad nothing to do with any measure.It remains to show that there exists a sequence of these measurable selections o,, suchthat the conclusion of 2.) holds. To do this define for a single @ € QPy (@) ={ P(@) 1B (¥n,2"1) fT (@) 1B (0,2 ") #0I'(@) otherwise when there is empty intersection *The following picture illustrates T,;(@) when @ is such that there is nonempty intersection.Also, given x € I'(@), and i, there is x, from the countable dense set such that the situationof the picture occurs.Is I',; measurable? If so, then from the above, it has a measurable selection 0,,; andthe set of these o,,; must have the property that {O,;(@)},,; is dense in '(@) for each @.Let U be open. Then{@ :Tyi(@) NU £0} = {@:T(@)NB(xn,2') NU ZO}U[{@:T(@) OB (%,2') =0}N{@:P(@) NU FO}= {@:T(@)NB(x%,2-') NU 4O}U[(Q\ {@ :T(@)NB (xn,2) 40}) N{w:T(@)NU 4 9}],a measurable set. Thus I’; is measurable as hoped.By what was just shown, there exists o,,;, a measurable function such that 0,;(@) €T,i(@) CT (@) for all @ € Q. Ifx ET (@), thenx €B (2n,.2-) whenever x, is closeenough to x. Thus both x,6,,¢;41) (@) are in B (x,,2~+))) and so |On(i+1) (@)—x|<27,It follows that condition 2.) holds with the countable dense subset of ['(@) being the{0,;(@)}. Note that this had nothing to do with a measure.Now consider why 2.)= 1.). We have {o,(@)} CT'(@) and o, is measurable andUnOn (@) equals (@). Why is I a measurable multifunction? Let U be an open setr-(U) = {@:T(@)NU 40} = {o:T(@)nu oh= U,o,|(U)eF &