29.2. CONDITIONAL EXPECTATION AND INDEPENDENCE 785

Proof: First I claim that σ (Z,Y ) consists of (Z,Y )−1 (B) where B is Borel in W ×WN

where WN ≡∏∞i=1 W . To see this, let{

(Z,Y )−1 (B) : B is Borel in W 1+N}≡F .

Then F is a σ algebra and for VJ of the form ∏∞j=1 Vj where Vj = W for all j except for

those contained in a finite set J and for the other j,Vj is an open set, and U an open setin W, then (Z,Y )−1 (U×VJ) = Z−1 (U)∩Y −1 (VJ) ∈F so that, in particular, choosingVJ and U appropriately shows Z,Yk are all measurable with respect to F . Hence F ⊇σ (Z,Y ). Also, U ×VJ just described, where U is in a countable basis for W and eachVj is W or in a countable basis for W is a countable basis for the topology of W 1+N. Bydefinition, σ (Z,Y ) must contain (Z,Y )−1 (U×VJ) for U ×VJ in this countable basis.Therefore, σ (Z,Y ) must contain (Z,Y )−1 (O) for all O open and so also σ (Z,Y ) mustcontain (Z,Y )−1 (B) for B Borel, so σ (Z,X) ⊇ F . Also, this shows that for K thesesets in the countable basis for W 1+N,σ (K ) = σ (Z,Y ) =F . Now consider the followingcomputation. ∫

(Z,Y )−1(U×VJ)E (X |σ (Z,Y ))dP

≡∫(Z,Y )−1(U×VJ)

XdP

=∫

XY −1(VJ)XZ−1(U)XdP ∗

= P(Y −1 (VJ)

)∫XZ−1(U)XdP

∗∗= P

(Y −1 (VJ)

)∫E(XZ−1(U)X |Z

)dP =

∫XY −1(VJ)

E(XZ−1(U)X |Z

)dP

=∫

Z−1(U)XY −1(VJ)

E (X |Z)dP =∫(Z,Y )−1(U×VJ)

E (X |Z)dP

Then ∗ happens because X[Z∈U ]X is E measurable and XY −1(VJ)is F measurable and by

assumption, these are independent σ algebras. ∗∗ happens because Ω is σ (Z) measurableand the definition of conditional expectation. Next step happens because XY −1(VJ)

is F

measurable and E(XZ−1(U)X |Z

)is E measurable. Then the rest follows because XZ−1(U)

is σ (Z) measurable so it comes out of the conditional expectation.Now let G be those sets G of σ (Z,Y ) = σ (K ) such that∫

GXdP≡

∫G

E (X |σ (Z,Y ))dP =∫

GE (X |σ (Z))dP≡

∫G

E (X |Z)dP

As just shown, K ⊆ G . If G = ∪∞k=1Gk where the Gk are disjoint and the equation holds

for each Gk∫G

E (X |Z)dP =∫ ∞

∑k=1

XGk E (X |Z)dP =∞

∑k=1

∫Gk

E (X |Z)dP

=∞

∑k=1

∫Gk

E (X |σ (Z,Y ))dP =∫ ∞

∑k=1

XGk E (X |σ (Z,Y ))dP

=∫

GE (X |σ (Z,Y ))dP

29.2. CONDITIONAL EXPECTATION AND INDEPENDENCE 785Proof: First I claim that o (Z,Y) consists of (Z,¥Y)~' (B) where B is Borel in W x WNwhere WN = J], W. To see this, let{(Z.¥) (B) : Bis Borel in W'tN \ =F.Then -¥ is a o algebra and for V; of the form []j_, Vj where V; = W for all j except forthose contained in a finite set J and for the other j,V; is an open set, and U an open setin W, then (Z,Y)~'(U x Vy) =Z~!(U) NY! (V,) € F so that, in particular, choosingV, and U appropriately shows Z,Y; are all measurable with respect to %. Hence F Do(Z,Y). Also, U x V; just described, where U is in a countable basis for W and eachV; is W or in a countable basis for W is a countable basis for the topology of Wwit\, Bydefinition, o(Z,Y) must contain (Z,Y)~'(U x V;) for U x V; in this countable basis.Therefore, 6 (Z,Y) must contain (Z,Y) | (O) for all O open and so also o(Z,Y) mustcontain (Z,Y)~' (B) for B Borel, so o(Z,X) D F. Also, this shows that for .% thesesets in the countable basis for W!*N ,o (.#) = 0 (Z,Y) = ¥. Now consider the followingcomputation./ E(X|o(Z,Y))dPJ(Z,¥)1(UxV;)Ly XdP¥)1(UxV5)= [0 z-1yyXdP = P(Y! (Vy) ) f %r y)XaP=P(Y"'(Vs)) I Cont dP = [Fert (271)XIZ) dP= bow )E (X|Z)dP = | E(X|Z)dP(Z,¥)!(Uxvy)Then * happens because 2izeu|X is & measurable and 2y- l(vy) is Y measurable and byassumption, these are independent o algebras. ** happens because Q is o (Z) measurableand the definition of conditional expectation. Next step happens because 24-1 (V;) is Fmeasurable and E (22-1w)X iz) is & measurable. Then the rest follows because 27-1 (U)is o (Z) measurable so it comes out of the conditional expectation.Now let Y be those sets G of o (Z, Y) = 0 (.#) such that[ xap= Leclozy)ar= | Ex\o@)aP= | B(x\z)aPAs just shown, # CY. If G=U_, Gx where the G; are disjoint and the equation holdsfor each G;[ear = [¥ %ek(x\z)ar= X [earL gE OZ) \)dP = [YL %e (X|o (Z,Y))dP| E(X|o(Z,Y))dPJG