221 Analysis 2, 2008–09 Summary of theorems and definitions Measure theory Definition. The extended real line is the set [−∞, ∞] = R ∪ {−∞, ∞} where −∞ and ∞ are two symbols which do not belong to R. We extend order, addition and product on R to [−∞, ∞] in the natural way by defining, for any x ∈ R: (i). −∞ < x, x < ∞ and −∞ < ∞; (ii). (−∞) + x = −∞ = x + (−∞), ∞ + x = ∞ = x + ∞, ∞ + ∞ = ∞ and (−∞) + (−∞) = −∞ (note that (−∞) + ∞ is not defined); if x > 0 ∞ (iii). ∞ · x = x · ∞ = 0 if x = 0 , ∞ · ∞ = ∞ = (−∞) · (−∞) and −∞ if x < 0 (−∞) · y = −(∞ · y) = y · (−∞) for any y ∈ [−∞, ∞]. Definition. Let X be a set. We write P(X) = {all subsets of X}. Definition. Let X be a set. A collection A ⊆ P(X) of subsets of X is called a ring of subsets of X , or simply a ring , if (i). ∅ ∈ A, (ii). A, B ∈ A =⇒ A \ B ∈ A, and (iii). A, B ∈ A =⇒ A ∪ B ∈ A. Definition. A measure is a function m : A → [0, ∞], where A is a ring of subsets of a set X, such that (i). m(∅) = 0, and ∞ ∞ [ X (ii). m Ai = m(Ai ) i=1 • Ai ∈ A for i ≥ 1, whenever i=1 • Ai ∩ Aj = ∅ for i 6= j, and S • ∞ i=1 Ai ∈ A. We abbreviate property (ii) by saying that m is countably additive on disjoint unions , or just countably additive . 1 Proposition. Let m : A → [0, ∞] be a measure, where A is a ring of subsets of X. Then (i). if A, B ∈ A with A ⊆ B, then m(A) ≤ m(B); (ii). if A, B ∈ A with A ⊆ B and m(B) < ∞ then m(B \ A) = m(B) − m(A). Lemma (The disjoint unions trick). If A is a ring of subsets of a set X ei ∈ A with and Ai ∈ A for i = 1, 2, . . . then there are pairwise disjoint sets A ei ⊆ Ai A ∞ [ and ei = A i=1 ∞ [ Ai . i=1 Theorem (Measures are countably subadditive). Let m : A → [0, ∞] be a measure where A is a ring. For any Ai ∈ A, i = 1, 2, . . . , we have ∞ ∞ [ X m Ai ≤ m(Ai ). i=1 i=1 Definition. If Ai ⊆ X for i = 1, 2, . . . and A ⊆ X, Ai ↑ A means A1 ⊆ A2 ⊆ . . . Ai ↓ A means A1 ⊇ A2 ⊇ . . . and A = and A = ∞ [ i=1 ∞ \ Ai ; Ai . i=1 Theorem (Increasing unions and decreasing intersections). Let m : A → [0, ∞] be a measure, where A is a ring of subsets of X, and let Ai ∈ A for i = 1, 2, . . . . (i). If Ai ↑ A where A ∈ A, then m(Ai ) → m(A) as i → ∞. (ii). If Ai ↓ A where A ∈ A and if m(A1 ) < ∞, then m(Ai ) → m(A) as i → ∞. Definition. Let X be a set. A collection A ⊆ P(X) of subsets of X is called a σ-algebra if (i). ∅ ∈ A, (ii). A ∈ A =⇒ A{ ∈ A [we write A{ = X \ A], and S (iii). Ai ∈ A for i = 1, 2, . . . =⇒ ∞ i=1 Ai ∈ A. Proposition. T If A is a σ-algebra and Ai ∈ A for i = 1, 2, . . . then ∞ i=1 Ai ∈ A. 2 Definition. Let m : A → [0, ∞] be a measure where A is a ring of subsets of a set X. The outer-measure of m is the function m∗ : P(X) → [0, ∞] defined for any Y ⊆ X by m∗ (Y ) = inf ∞ nX m(Ai ) : Ai ∈ A for i = 1, 2, . . . and Y ⊆ i=1 ∞ [ o Ai . i=1 Theorem. If A ∈ A then m∗ (A) = m(A). So m∗ extends m. Theorem. (i). If A, B ⊆ X with A ⊆ B then m∗ (A) ≤ m∗ (B). (ii). m∗ is countably subadditive. That is, if Yi ⊆ X for i = 1, 2, . . . then m ∗ ∞ [ Yi ≤ i=1 ∞ X m∗ (Yi ). i=1 Definition. Let m : A → [0, ∞] be a measure where A is a ring of subsets of a set X. A set B ⊆ X is m-measurable if for any Y ⊆ X, m∗ (Y ) = m∗ (Y ∩ B) + m∗ (Y \ B). no proof Theorem (The Extension Theorem). Let m : A → [0, ∞] be a measure where A is a ring of subsets of a set X, and let M denote the collection of all m-measurable subsets of X. (i). A ⊆ M, (ii). M is a σ-algebra, and (iii). m∗ |M : M → [0, ∞] is a measure which extends m. Corollary. For any measure m : A → [0, ∞], there is an extension of m to a measure on the σ-algebra M of all m-measurable sets. Definition. Let m : A → [0, ∞] be a measure where A is a ring of subsets of a set X. We say that m is σ-finite if there are sets Xi ∈ A with Xi ↑ X such that m(Xi ) < ∞ for i = 1, 2, . . . . Theorem (Uniqueness of extensions for σ-finite measures). For any σ-finite measure m : A → [0, ∞], there is a unique extension of m to a measure on the σ-algebra M of all m-measurable sets. 3 Lebesgue measure on R Definition. The interval ring is the set I ⊆ P(R), I = {all finite unions of intervals of the form (a, b] where a, b ∈ R}. Definition. The length measure m : I → [0, ∞] is defined by m((a1 , b1 ] ∪ (a2 , b2 ] ∪ · · · ∪ (an , bn ]) = b1 − a1 + b2 − a2 + . . . + bn − an where a1 < b1 < a2 < b2 < · · · < an < bn . no proof Theorem. The length measure m : I → [0, ∞] is a (well-defined) σ-finite measure. Definition. Let m : I → [0, ∞] be the length measure. A set B ⊆ R is Lebesgue measurable if it is m-measurable. Let M denote the collection of all Lebesgue measurable sets. Lebesgue measure m : M → [0, ∞] is the measure obtained by extending the length measure to the σ-algebra M of Lebesgue measurable sets using the Extension Theorem. For the rest of this section, we will write m for Lebesgue measure and M for the collection of all Lebesgue measurable sets. Proposition. Let A ∈ M and c ∈ R. (i). m(A + c) = m(A) [m is translation invariant], and (ii). m(cA) = |c|m(A). no proof Theorem. There is a subset of R which is not Lebesgue measurable. Definition. Let S ⊆ P(R) be any collection of subsets of R. The σ-algebra generated by S is o \n σ(S) = A ⊆ P(R) : A is a σ-algebra and S ⊆ A Proposition. σ(S) is the smallest σ-algebra of subsets of R which contains S. That is, (i). σ(S) is a σ-algebra and S ⊆ σ(S), and (ii). If B ⊆ P(R) is any σ-algebra with S ⊆ B then σ(S) ⊆ B. Definition. The collection of Borel subsets of R is defined to be σ(O) where O is the collection of open subsets of R. Theorem. The collection of Borel subsets of R is equal to σ(I). So I ⊆ σ(I) = σ(O) ⊆ M ( P(R). Theorem. For any Lebesgue measurable set A ∈ M, there is a Borel set B with A ⊆ B and m(B \ A) = 0. 4 Integration with respect to a measure Definition. A measure space consists of a set X, a σ-algebra M ⊆ P(X) and a measure m : M → [0, ∞]. Let us fix a measure space (X, M, m). We refer to the sets in M as measurable sets. Definition. A function f : X → [−∞, ∞] is measurable if the set f −1 (α, ∞] = {x ∈ X : f (x) > α} is in M for every α ∈ R. Definition. If A ⊆ X then the characteristic function of A is the function χA : X → R defined by ( 1 if x ∈ A, χA (x) = 0 if x 6∈ A. Proposition. If f, g : X → [−∞, ∞] are two measurable functions, A ∈ M is a measurable set and c ∈ [−∞, ∞], then f + g, f χA and cf are all measurable functions. Definition. A function ϕ : X → R is simple if the range of ϕ is a finite set. Proposition. Let ϕ : X → R be a simple function. There is a partition X = A1 ∪ · · · ∪ An of X and a1 , . . . , an ∈ R with ai 6= aj if i 6= j such that ϕ = a1 χ1 + · · · + an χn . (F) Apart from re-ordering, there is only one way to write ϕ in this form. Moreover, ϕ is measurable if and only if each set A1 , . . . , An is measurable. Definition. If ϕ : X → [0, ∞) is a nonnegative, simple measurable function written in the form (F) then the integral of ϕ with respect to m is defined as Z ϕ dm = a1 m(A1 ) + · · · + an m(An ). If A is a measurable set then ϕχA is a nonnegative, simple measurable function, so we can define the integral of ϕ over A with respect to m by Z Z ϕ dm = ϕχA dm. A Proposition. If A is any measurable set then 5 R χA dm = m(A). Proposition. If c ∈ [0, ∞) and ϕ, ψ : X → [0, ∞) are nonnegative, simple measurable functions then (i). Rϕ + ψ is nonnegative, simple and measurable, and R R ϕ + ψ dm = ϕ dm + ψ dm, and R R (ii). cϕ is nonnegative, simple and measurable, and cϕ dm = c ϕ dm. Theorem. If ϕ : X → [0, ∞) is a nonnegative, simple measurable function then the map λ : M → [0, ∞], Z λ(A) = ϕ dm, for A ∈ M A is a measure. Corollary. Let ϕ : X → [0, ∞) be a nonnegative, simple measurable function. Let A ∈ M and An ∈ M for n = 1, 2, . . . , with An ↑ A. Then Z Z ϕ dm → ϕ dm as n → ∞. An A Definition. Let f : X → [0, ∞] be a nonnegative measurable function. The integral of f with respect to m is defined as Z nZ o ϕ : X → [0, ∞) is a nonnegative, simple f dm = sup ϕ dm : . measurable function with 0 ≤ ϕ ≤ f If A is a measurable set then f χA is a nonnegative measurable function, so we can define the integral of f over A with respect to m by Z Z f dm = f χA dm. A Proposition. Let f and g be nonnegative measurable functions X → [0, ∞] and let A, B ∈ M. R R (i). If f ≤ g then f dm ≤ g dm. R R (ii). If A ⊆ B then A f dm ≤ B f dm. Definition. A sequence (fn )n≥1 of functions fn : X → [−∞, ∞] is monotone increasing if f1 (x) ≤ f2 (x) ≤ f3 (x) ≤ . . . for every x ∈ X. If f : X → [−∞, ∞], we write f = lim fn or fn → f as n → ∞ if n→∞ fn (x) → f (x) for every x ∈ X. Proposition. Let (fn )n≥1 be a monotone increasing sequence of functions fn : X → [−∞, ∞]. (i). There is a function f : X → [−∞, ∞] with f = lim fn . n→∞ (ii). If each fn is a measurable function, then so is f = lim fn . n→∞ 6 Theorem (The Monotone Convergence Theorem). If (fn )n≥1 is a monotone increasing sequence of nonnegative measurable functions fn : X → [0, ∞] then Z Z lim fn dm = lim fn dm. n→∞ n→∞ Theorem. Let f : X → [0, ∞] be any nonnegative measurable function. There is a monotone increasing sequence (ϕn )n≥1 of nonnegative, simple measurable functions ϕn : X → [0, ∞) with f = lim ϕn . n→∞ Corollary. (i). If f, g : X → [0, ∞] are nonnegative measurable functions and c ≥ 0 then f + g and cf are nonnegative measurable functions, and Z Z Z Z Z f + g dm = f dm + g dm and cf dm = c f dm. (ii). If fn :P X → [0, ∞] are nonnegative measurable functions for n ≥ 1 then ∞ n=1 fn is a nonnegative measurable function and Z X ∞ fn dm = ∞ Z X fn dm. n=1 n=1 Corollary. If f : X → [0, ∞) is a nonnegative measurable function then the map λ : M → [0, ∞], Z f dm, for A ∈ M λ(A) = A is a measure. Corollary. Let f : X → [0, ∞) be a nonnegative measurable function. Let Ai ∈ M for i = 1, 2, . . . . S (i). If A = ∞ i=1 Ai is a countable disjoint union then Z f dm = A (ii). If Ai ↑ A then R f dm → ∞ Z X i=1 R f dm. Ai f dm as i → ∞. R R R (iii). If Ai ↓ A and A1 f dm < ∞ then Ai f dm → A f dm as i → ∞. Ai A 7 Definition. A set A ∈ M is m-null , or null , if m(A) = 0. Definition. Let f, g : X → [−∞, ∞] be measurable functions. We say that f = g almost everywhere , or f = g a.e. if the set {x ∈ X : f (x) 6= g(x)} is m-null. Theorem. Let f : X → [0, ∞] be a nonnegative measurable function. Then Z f dm = 0 ⇐⇒ f = 0 almost everywhere. Corollary. If A is an m-null set and f : X → [0, ∞] is any nonnegative R measurable function then A f dm = 0. Corollary. If f, g : X → [0, ∞]Rare nonnegative measurable functions with R f = g almost everywhere, then f dm = g dm. Theorem (An almost everywhere version of the MCT). If (fn )n≥1 is a monotone increasing sequence of nonnegative measurable functions fn : X → [0, ∞] with fn ↑ f almost everywhere then Z Z f dm = lim fn dm. n→∞ Definition. If f : X → R is a measurable function, we say f is integrable with respect to m if Z Z + f dm < ∞ and f − dm < ∞ where f + , f − : X → [0, ∞) are the nonnegative measurable functions ( ( f (x) if f (x) ≥ 0 −f (x) if f (x) ≤ 0 f + (x) = and f − (x) = 0 if f (x) < 0 0 if f (x) > 0 If f is integrable with respect to m, we define the integral of f with respect to m by Z Z Z + f dm = f dm − f − dm. Proposition. If fR1 , f2 : X → [0, ∞) are measurable functions with R f1 dm < ∞ and f2 dm < ∞ then the function f = f1 − f2 is integrable and Z Z Z f dm = f1 dm − f2 dm. Corollary. If f, g : X → R are integrable functions and c ∈ R then f + g and cf are integrable with Z Z Z Z Z f + g dm = f dm + g dm and cf dm = c f dm. 8 Theorem. If f, gR : X → RRare integrable functions with f = g almost everywhere then f dm = g dm. Theorem. If f : X → R is measurable then |f | is measurable and (i). f is integrable ⇐⇒ |f | is integrable; and R R (ii). if f is integrable then | f dm| ≤ |f | dm. Definition. A complex-valued function f : X → C is integrable if the real-valued functions Re f and Im f are both integrable. If f is integrable, the integral of f with respect to m is the complex number Z Z Z f dm = Re f dm + i Im f dm. R R Theorem. If f : X → C is integrable then | f dm| ≤ |f | dm. Definition. Let (an )n≥1 be a sequence in [−∞, ∞]. The lim sup and lim inf of this sequence are defined by lim sup an = lim sup ai and n→∞ n→∞ lim inf an = lim n→∞ i≥n n→∞ inf ai . i≥n Theorem (Fatou’s lemma). If (fn )n≥1 is a sequence of nonnegative measurable functions fn : X → [0, ∞] then Z Z lim inf fn dm ≤ lim inf fn dm. n→∞ n→∞ Theorem (The Dominated Convergence Theorem). Suppose that (fn )n≥1 is a sequence of integrable functions fn : X → R and that fn → f as n → ∞. If there is an integrable function g : X → R with |fn | ≤ g for all n ≥ 1, then Z Z f is integrable, and f dm = lim fn dm. n→∞ Theorem. functions fn : X → R so P∞ Let (fn )n≥1 be a sequence of integrableP ∞ R that n=1 |fn (x)| < ∞ for every x ∈ X and with n=1 |fn | dm < ∞. Then Z X ∞ Z ∞ X fn dm. fn dm = n=1 n=1 Theorem (Differentiation under the integral sign). Let f : X × [a, b] → R be a function so that (i). for each t ∈ [a, b], x 7→ f (x, t) is an integrable function X → R; and (ii). for each x ∈ X, t 7→ f (x, t) is a differentiable function [a, b] → R; and (iii). for some integrable g : X → R we have ∂f (x, t) ≤ g(x) ∂t for every t ∈ [a, b] and x ∈ X. Then d dt Z Z f (x, t) dx = 9 ∂f (x, t) dx ∂t Lebesgue integration on R Definition. The Lebesgue integral is the integral with respect to Lebesgue measure m : M → R where M is the collection of Lebesgue measurable subsets of R. In other words, integrals with respect to m are defined by applying the preceding theory to the measure space (R, M, m) where M and m are the Lebesgue measurable subsets of R, and Lebesgue measure on R, respectively. For a ≤ b, we’ll write Z Z a Z b f (x) dx instead of f dm f (x) dx = − b a (a,b) Proposition. For any a, b, c ∈ R and any integrable function f : R → R, Z c Z c Z b f (x) dx + f (x) dx = f (x) dx. a b a Theorem (The Fundamental Theorem of R t Calculus). Let a ∈ R, let f : R → R be continuous and let F (t) = a f (x) dx for t ∈ R. Then F 0 (t) = f (t) for all t ∈ R. Corollary. If G : R → R is continuously differentiable and a, b ∈ R then Rb 0 G (x) dx = G(b) − G(a). a u f Corollary. If a, b, c, d ∈ R and u and f are functions [a, b] → [c, d] → R with u continuously differentiable and f continuous, then Z b Z 0 u(b) f (u(x)) u (x) dx = a f (y) dy. u(a) Theorem. Let f : R → R be an integrable function and let c ∈ R. R R (i). f (x + c) dx = f (x) dx R R 1 (ii). If c 6= 0 then f (cx) dx = |c| f (x) dx. 10 Multiple integration Let (X, L, `) and (Y, M, m) be measure spaces. We will write Z Z Z Z f (x) d`(x) = f d` and g(y) dm(y) = g dm whenever the expressions on the right hand side make sense. Definition. If A ∈ L and B ∈ M then the set A × B is called a rectangle . We write rect(L, M) = {A × B : A ∈ L, B ∈ M} for the set of all rectangles. We also write Arect (L, M) = {all finite unions of rectangles A × B ∈ rect(L, M)}. Proposition. Arect (L, M) is a ring of subsets of X × Y , and every element of Arect (L, M) is a finite disjoint union of rectangles from rect(L, M). S Definition. If E ∈ Arect (L, M) with E = ∞ i=1 Ai × Bi is a disjoint union (possibly with Ai × Bi = ∅ for i sufficiently large) then we define X π(E) = `(Ai )m(Bi ). i≥1 Theorem. π : Arect (L, M) → [0, ∞] is a well-defined measure. Definition. The product σ-algebra of the σ-algebras L and M is L × M = σ(Arect (L, M)). The product measure of the measures ` and m is l × m = π ∗ |L×M , the restriction of the outer measure π ∗ to L × M. Since L × M is contained in the π-measurable subsets of X × Y , the Extension Theorem shows that ` × m is a measure. Monotone classes Definition. Let X be a set. A monotone class of subsets of X is a non-empty collection C ⊆ P(X) which is closed under increasing unions and decreasing intersections: (i). if An ∈ C and An ↑ A for some A ⊆ X, then A ∈ C; and (ii). if An ∈ C then and An ↓ A for some A ⊆ X then A ∈ C. If S ⊆ P(X) is a non-empty collection of subsets of X, we write mon(S) for the smallest monotone class of subsets of X which contains S. 11 Proposition. (i). A non-empty collection of subsets of X which is closed under increasing unions and complements is a monotone class. (ii). Any σ-algebra is a monotone class. (iii). S ⊆ mon(S) ⊆ σ(S) for any non-empty collection S of subsets of X. no proof Theorem (Monotone class lemma). If A ⊆ P(X) is a ring of subsets of X then mon(A) is also a ring of subsets of X. Corollary. Let A be a ring of subsets of a set X. If X ∈ mon(A) then mon(A) = σ(A). Corollary. L × M = mon(Arect (L, M)). Integration using product measure Definition. Let E ⊆ X × Y . For x ∈ X and y ∈ Y , we write Ex = {y ∈ Y : (x, y) ∈ E} and E y = {x ∈ X : (x, y) ∈ E}. Theorem. If E ∈ L × M then Ex ∈ M for every x ∈ X, and E y ∈ L for every y ∈ Y . no proof Theorem. Let l and m be σ-finite measures and let E ∈ L × M. (i). Both of the functions X → [0, ∞], x 7→ m(Ex ) and Y → [0, ∞], y 7→ m(E y ) are measurable. Z Z (ii). l × m(E) = m(Ex ) dl(x) = l(E y ) dm(y). X Y Theorem (Tonelli’s theorem). If F : X × Y → [0, ∞] is a nonnegative measurable function then Z Z Z Z Z F (x, y) dm(y) dl(x) = F d(l×m) = F (x, y) dl(y) dm(x). X Y X×Y Y X [In particular, all of the functions that must be measurable for these integrals to be defined, are measurable!] Theorem (Fubini’s theorem). If F : X × Y → R is integrable then Z Z Z Z Z F (x, y) dm(y) dl(x) = F d(l×m) = F (x, y) dl(y) dm(x). X Y X×Y Y R X Here, the Rintegral Y F (x, y) dm(y) is defined for almost every x ∈ X, and x 7→ Y F (x, y) dm(y) R is integrable. Similarly,R the integral X F (x, y) dl(x) is defined for almost every y ∈ Y , and y 7→ X F (x, y) dl(x) is integrable. 12 Lebesgue measure and integration on Rn Definition. Let (R, M, m) be the usual Lebesgue measure space, as on page 10. Lebesgue measure on R2 is the measure m × m : M × M → [0, ∞]. We obtain a corresponding integral from the measure space (R2 , M × M, m × m). Similarly, we define Lebesgue measure on Rn by m × · · · × m : M × · · · × M → [0, ∞], and obtain a corresponding integral from the measure space (Rn , M × · · · × M, m × · · · × m). If f : Rn → R is integrable then we write Z f (x) dx Rn for the integral of f with respect to Lebesgue measure on Rn . We have the following generalisation of the final theorem on page 10: Theorem. Let A be an invertible n × n matrix with real entries. If f : Rn → R is integrable then Z Z 1 f (x) dx. f (Ax) dx = | det A| Rn Rn Corollary. Let A be any n × n matrix with real entries and let µ = m × · · · × m denote Lebesgue measure on Rn . If S ⊆ Rn is a Lebesgue measurable and we write A(S) = {Ax : x ∈ S} then µ(A(S)) = | det A|µ(S). 13