Product Ο-algebra: Ο-algebra generated by the measurable rectangles A Γ B with A β Eβ and B β Eβ.
x-section: For C β Eβ Γ Eβ and x β Eβ, the x-section of C is Cβ = {y β Eβ : (x, y) β C}.
β Must-know
If C belongs to Eβ β Eβ, then Cβ belongs to Eβ for every x β Eβ, and CΚΈ belongs to Eβ for every y β Eβ.
Further detail
For a finite collection of measurable spaces (Eα΅’, Eα΅’), the product Ο-algebra on Eβ Γ β― Γ Eβ is generated by the rectangles Aβ Γ β― Γ Aβ with Aα΅’ β Eα΅’.
π 2. Measurable Sections and Sigma-Finiteness
π Key Concepts & Definitions
Ο-finite measure: a measure on (E, E) for which there is an increasing sequence of measurable sets (Xβ) such that ββ Xβ = E and ΞΌ(Xβ) < β for every n.
Ο-finite measure: A measure ΞΌ on a measurable space (E, π) is Ο-finite if there are measurable sets Xβ increasing to E and satisfying ΞΌ(Xβ) < β for every n.
π Essential Points
If f : (Eβ Γ Eβ, Eβ β Eβ) β (G, G) is measurable, then for every x β Eβ the function fβ(y) = f(x, y) is Eβ-measurable, and for every y β Eβ the function fΚΈ(x) = f(x, y) is Eβ-measurable.
π 3. Product measure
π Key Concepts & Definitions
Product Measure: The measure m on E1βΓE2β is defined by m(C)=β«E1ββΟ(Cxβ)ΞΌ(dx) for measurable sets C.
β Must-know
If ΞΌ and Ξ½ are Ο-finite measures on (Eβ, πβ) and (Eβ, πβ), respectively, there exists a unique measure ΞΌ β Ξ½ on (Eβ Γ Eβ, πβ β πβ) such that ΞΌ β Ξ½(A Γ B) = ΞΌ(A)Ξ½(B) for every A β πβ and B β πβ, with the convention 0 Γ β = 0.
π Formula - For every C β πβ β πβ, ΞΌβΞ½(C)=β«E1ββΞ½(Cxβ)ΞΌ(dx)=β«E2ββΞΌ(Cyβ)Ξ½(dy).
π Formula - For measurable rectangles, the product measure satisfies m(AΓB)=ΞΌ(A)Ο(B) for AβE1β and BβE2β.
Further detail
The product measure ΞΌ β Ξ½ of two Ο-finite measures is Ο-finite.
For every C β πβ β πβ, the section measure x β¦ Ξ½(Cβ) is πβ-measurable.
The construction of m is completed by proving that it is a measure; this proof is left as an exercise.
π 4. Independence of Random Variables
π Key Concepts & Definitions
Independence of Random Variables: Random variables X1β,β¦,Xnβ are independent if, for every choice of measurable sets BkββEkβ and every subset Jβ{1,β¦,n}, P(βkβJβ{XkββBkβ})=βkβJβP(XkββBkβ).
Independence of Events: Events A1β,β¦,Anβ are independent if, for every subset Jβ{1,β¦,n}, P(βkβJβAkβ)=βkβJβP(Akβ).
Joint Law: The law of (X,Y) is the probability measure defined on measurable sets C by P(X,Y)β(C)=P((X,Y)βC).
β Must-know
π Two random variables X and Y are independent if and only if their joint law satisfies P(X,Y)β=PXββPYβ.
Further detail
A pair-valued map (X,Y) is measurable into the product measurable space if and only if X and Y are measurable into their respective spaces.
For finitely many random variables, independence is equivalent to the joint law being the product of their individual laws, and the product of measures is associative.
π‘ Memory Hook
Event independence concerns events; random-variable independence concerns their inverse-image events.
π 5. Integration Against Product Measures
β Must-know
π Formula - For a nonnegative simple function f=βk=1nβΞ±kβ1Akββ on a product space, its integral against ΞΌ is β«fdΞΌ=βk=1nβΞ±kβΞΌ(Akβ), where the sets Akβ are measurable and the coefficients Ξ±kβ are nonnegative.
π Formula - For a nonnegative measurable function f, its integral against ΞΌ is β«fdΞΌ=sup{β«gdΞΌ:gΒ isΒ simpleΒ andΒ gβ€f}.
π A measurable function f belongs to L1 for the measure ΞΌ exactly when β«β£fβ£dΞΌ<+β.
Further detail
π Formula - For an integrable measurable function f, its integral is defined by β«fdΞΌ=β«f+dΞΌββ«fβdΞΌ, where f+=max(f,0) and fβ=βmin(f,0).
π‘ Memory Hook
Simple functions first, nonnegative limits next, signed integrable functions last.
π 6. Lebesgue Measure in Dimension d
π Key Concepts & Definitions
Volume of a d-cell: The volume of the d-cell W=βk=1dβ]akβ,bkβ[ is Vol(W)=βk=1dβ(bkββakβ) when akβ<bkβ for every coordinate.
β Must-know
The product measure Ξ»βd is the unique measure satisfying Ξ»βd(A1βΓβ―ΓAdβ)=βk=1dβΞ»(Akβ) for Borel sets A1β,β¦,Adβ.
The Lebesgue measure on Rd is the restriction of the outer measure to its measurable sigma-algebra, which contains the Borel sigma-algebra and is its completion with respect to Lebesgue measure.
π Lebesgue measure on Rd is the unique measure on the Borel sigma-algebra that assigns each d-cell its volume.
Further detail
The Borel sigma-algebra on Rd is the product of d copies of the Borel sigma-algebra on R.
π Formula - For any subset AβRd, the outer measure is Ξ»(d,β)(A)=inf{βjβVol(Wjβ):AββjβWjβ,Β WjβΒ areΒ d-cells}.
The d-fold product of one-dimensional Lebesgue measure equals Lebesgue measure on Rd.
π 7. Change of Variables Formula
π Key Concepts & Definitions
C1-diffeomorphism: A map Ο:VβU between open subsets of Rd is a C1-diffeomorphism if it is bijective and both it and its inverse are differentiable with continuous differentials.
Jacobian determinant: The Jacobian determinant of a differentiable map Ο:VβRd is JΟβ(x)=det(βxiββΟjββ(x))i,jβ.
β Must-know
π For a C1-diffeomorphism Ο:VβU and measurable f:UβR, f is integrable on U if and only if β£JΟββ£(fβΟ) is integrable on V.
π Formula - For a CΒΉ-diffeomorphism Ο:VβU and an integrable function f on U, β«Uβf(y)Ο(dy)=β«Vββ£JΟ(x)β£f(Ο(x))Ο(dx), and the equality also holds for measurable nonnegative f with values in [0,+β].
π For a CΒΉ-diffeomorphism Ο:VβU, f is integrable on U if and only if |JΟ|(fβΟ) is integrable on V.
Further detail
π The change-of-variables integral equality also holds with values in [0,+β] for every measurable nonnegative function on U.
The change-of-variables proof extends the result from indicator functions to simple functions, then to nonnegative measurable functions, and finally to integrable functions.
The admitted first step of the change-of-variables proof states that for any cuboid CβV, Ο(Ο(C))=β«_C |JΟ| dΟ.
π 8. Fubini Theorems
β Must-know
π For Ο-finite measures ΞΌ and Ξ΅ and a measurable function f:EβΓEββ[0,+β], the functions xβ¦β«{Eβ}f(x,y)Ξ΅(dy) and yβ¦β«{Eβ}f(x,y)ΞΌ(dx) are measurable.
π Formula - For Ο-finite measures ΞΌ and Ξ΅ and measurable f:EβΓEββ[0,+β], β«E1βΓE2ββfd(ΞΌβΞ΅)=β«E1ββ(β«E2ββf(x,y)Ξ΅(dy))ΞΌ(dx)=β«E2ββ(β«E1ββf(x,y)ΞΌ(dx))Ξ΅(dy).
π For f integrable with respect to ΞΌβΞ΅, its sections yβ¦f(x,y) are integrable for ΞΌ-almost every x, and its sections xβ¦f(x,y) are integrable for Ξ΅-almost every y.
Further detail
For integrable f with respect to ΞΌβΞ΅, the two iterated integrals are well-defined except possibly on null sets and define integrable functions on their respective spaces.
π Formula - For f integrable with respect to ΞΌβΞ΅, β«E1βΓE2ββfd(ΞΌβΞ΅)=β«E1ββ(β«E2ββf(x,y)Ξ΅(dy))ΞΌ(dx)=β«E2ββ(β«E1ββf(x,y)ΞΌ(dx))Ξ΅(dy).
Random variables X and Y are independent if and only if E[f(X)g(Y)]=E[f(X)]E[g(Y)] for every pair of nonnegative measurable functions f and g, or equivalently for every pair of bounded measurable functions f and g.
For measurable non-negative bounded functions f and g, independence of X and Y implies E[f(X)g(Y)] = E[f(X)]E[g(Y)].
If E[f(X)g(Y)] = E[f(X)]E[g(Y)] for all measurable non-negative bounded f and g, then X and Y are independent, as shown by taking f and g to be indicator functions of measurable sets.
Further detail
π Formula - For independent Eβ-valued random variables Xβ,β¦,Xβ and nonnegative measurable functions fβ, E[βk=1nβfkβ(Xkβ)]=βk=1nβE[fkβ(Xkβ)]; the identity also holds when all fβ are bounded and measurable.
For independent random variables X and Y, the factorization of expectations also holds for bounded measurable complex-valued functions f and g.
The equivalence between independence and factorization of expectations extends to n random variables by applying Fubini nβ1 times.
π The components Xβ, β¦, X_d of a random vector are independent if and only if Ο_X(t) = β{k=1}^d Ο{X_k}(t_k) for every t = (tβ, β¦, t_d) β βα΅.
Joint cumulative distribution function: The joint cumulative distribution function of real-valued random variables X and Y is F_{(X,Y)}(x,y) = P(X β€ x, Y β€ y), for (x,y) β βΒ².
β Must-know
π Real-valued random variables X and Y are independent if and only if F_{(X,Y)}(x,y) = F_X(x)F_Y(y) for every (x,y) β βΒ².
π When the joint law has density f_{(X,Y)}, X and Y are independent if and only if f_{(X,Y)}(x,y) = f_X(x)f_Y(y) for Lebesgue-almost every (x,y) β βΒ².
Further detail
If the joint law of (X,Y) has density f_{(X,Y)} with respect to Lebesgue measure on βΒ², then its marginal densities are f_X(x) = β« f_{(X,Y)}(x,y) dy and f_Y(y) = β« f_{(X,Y)}(x,y) dx, and these integrals are finite outside a Lebesgue-null set.
π‘ Memory Hook
CDF factorization tests independence directly; density factorization does so almost everywhere.
Test yourself
Take a 30-question practice quiz on Product Integration and Change of Variables, with every answer explained.
1. How is the product Ο-algebra on E1βΓE2β defined?
2. Suppose CβE1ββE2β. What can be concluded about its sections for each fixed xβE1β and yβE2β?