Real Analysis Chapter 1 Solutions Jonathan Conder 3. (a) Let M be an infinite σ-algebra of subsets of some set X. There exists a countably infinite subcollection C ⊆ M, and we may choose C to be closed under taking complements (adding in missing complements if necessary). For each x ∈ X, define Dx := ∩{C ∈ C | x ∈ C}, so that Dx ∈ M. Let x, y ∈ X and suppose y ∈ Dx . Then y ∈ C for all C ∈ C with x ∈ C. Moreover, if y ∈ C for some C ∈ C, then y ∈ / C c ∈ C, so x ∈ / C c and hence x ∈ C. This implies that, if C ∈ C, then x ∈ C iff y ∈ C. In particular Dx = Dy . If x, y ∈ X and there exists z ∈ Dx ∩ Dy , then Dx = Dz = Dy , so the collection D := {Dx | x ∈ X} is pairwise disjoint. If C ∈ C, then C = ∪{Dx | x ∈ C}, since x ∈ Dx ⊆ C for all x ∈ C. Therefore, the image of the map ∪ : 2D → 2X contains C, so there exists a surjection from some subset of 2D onto C. If D were finite, then 2D would be as well, which is a contradiction because C is infinite. Therefore D is an infinite, pairwise disjoint subcollection of M. (b) Let M be an infinite σ-algebra of subsets of some set X. By the previous exercise, there exists an infinite, pairwise disjoint subcollection D of M. Since Q is countable, there exists an injection q : Q → D. Define r : 2Q → M by r(A) := ∪a∈A q(a). Then r is well-defined because each A ⊆ Q is countable, and injective because D is pairwise disjoint. There exists an injection from R → 2Q , for example the map x 7→ (−∞, x] ∩ Q (which is injective by the least upper bound property of R). Composing these injections gives an injective map from R → M, which shows that card(M) ≥ c. 4. Let A be an algebra which is closed under countable increasing unions. To show that A is a σ-algebra, it suffices to show that it is closed under arbitrary countable unions. To this end, let {En }n∈N be a countable collection of members of A. For each n ∈ N define Fn := ∪nk=1 Ek . Since A is an algebra it is clear that (Fn )n∈N is an increasing sequence of members of A, so ∪n∈N En = ∪n∈N Fn ∈ A as required. Conversely, it is plain that every σ-algebra is closed under countable increasing unions. 5. Let E be a collection of subsets of some set X, and let M be the σ-algebra generated by E. Define N := ∪{A | A is the σ-algebra generated by some countable subcollection of E}. It is clear that ∅, X ∈ N, E ⊆ N and N is closed under taking complements. Let {En }n∈N be a countable collection of members of N. For each n ∈ N there exists a countable subcollection En of E such that En is in the σ-algebra generated by En . Since union of countably many countable sets is countable, F := ∪n∈N En is a countable subcollection of E. Let A be the σ-algebra generated by F, so that A ⊆ N. For each n ∈ N the σ-algebra generated by En is a subset of A, because A is a σ-algebra containing En . This implies that En ∈ A for all n ∈ N, and hence ∪n∈N En ∈ A ⊆ N. Therefore N is closed under countable unions, so it is a σ-algebra containing E. This implies that M ⊆ N. Conversely, let E ∈ N. Then E belongs to the σ-algebra A generated by some countable subcollection of E. Since M is a σ-algebra containing this subcollection, A ⊆ M and hence E ∈ M. This shows that N ⊆ M and hence N = M. 6. Note that µ(∅) = µ(∅ ∪ ∅) = µ(∅) = 0. Given a sequence (An )n∈N of disjoint sets in M, there exists a sequence (En )n∈N of sets in M and a sequence (Fn )n∈N of subsets of measure zero sets from M such that An = En ∪ Fn for all n ∈ N. Note that (En )n∈N is pairwise disjoint, and ∪n∈N Fn is a subset of a measure zero set. Therefore X X µ(∪n∈N An ) = µ((∪n∈N En ) ∪ (∪n∈N Fn )) = µ(∪n∈N En ) = µ(En ) = µ(An ). n∈N n∈N This shows that µ is a measure. Let A ⊆ X, and suppose there exists B ∈ M such that A ⊆ B and µ(B) = 0. Then B = E ∪ F for some E ∈ M and F a subset of a measure zero set N ∈ M. It follows that A ⊆ E ∪ N, where E ∪ N ∈ M and µ(E ∪ N ) ≤ µ(E) + µ(N ) = µ(E) + 0 ≤ µ(B) = 0. Therefore A = ∅ ∪ A ∈ M, which shows that µ is complete. 1 Real Analysis Chapter 1 Solutions Jonathan Conder Let λ : M → [0, ∞] be another measure which extends µ, and let A ∈ M. Then A = E ∪ F for some E ∈ M and F a subset of a measure zero set N ∈ M. It follows that µ(E) = λ(E) ≤ λ(A) ≤ λ(E) + λ(F ) ≤ λ(E) + λ(N ) = µ(E) + µ(N ) = µ(E). Therefore λ(A) = µ(E) = µ(A), so λ = µ. This shows that µ is the unique measure on M which extends µ. 7. Set µ := Pn j=1 aj µj , and note that µ(∅) = 0. If {Ei }i∈N is a pairwise disjoint collection of members of M, then µ(∪i∈N Ei ) = n X j=1 aj µj (∪i∈N Ei ) = n X aj j=1 X µj (Ei ) = n XX aj µj (Ei ) = i∈N j=1 i∈N X µ(Ei ) i∈N (note that we are allowed to change the order of summation because all the terms are non-negative; alternatively you can consider the convergent/divergent cases separately). This shows that µ is a measure on (X, M). 8. Let (X, M, µ) be a measure space and {En }n∈N a countable collection of measurable sets. Then (∩∞ k=n Ek )n∈N is an ∞ increasing sequence of measurable sets with µ(∩k=n Ek ) ≤ µ(En ) for all n ∈ N, so ∞ ∞ µ(lim inf En ) = µ(∪n∈N (∩∞ k=n Ek )) = lim µ(∩k=n Ek ) = lim inf µ(∩k=n Ek ) ≤ lim inf µ(En ). n→∞ n→∞ n→∞ n∈N ∞ Moreover (∪∞ k=n Ek )n∈N is a decreasing sequence of measurable sets with µ(∪k=n Ek ) ≥ µ(En ), so ∞ ∞ µ(lim sup En ) = µ(∩n∈N (∪∞ k=n Ek )) = lim µ(∪k=n Ek ) = lim sup µ(∪k=n Ek ) ≥ lim sup µ(En ) n→∞ n∈N n→∞ n→∞ provided that the first term of the sequence has finite measure, i.e. µ(∪∞ k=1 Ek ) < ∞. 9. Let (X, M, µ) be a measure space and E, F ∈ M. Then E = (E ∩ F ) t (E \ F ) and (E \ F ) t F = E ∪ F, so µ(E) + µ(F ) = µ(E ∩ F ) + µ(E \ F ) + µ(F ) = µ(E ∩ F ) + µ(E ∪ F ). 10. Clearly µE (∅) = µ(∅ ∩ E) = µ(∅) = 0. If {En }n∈N is a pairwise disjoint collection of members of M, then X X µE (∪n∈N En ) = µ((∪n∈N En ) ∩ E) = µ(∪n∈N (En ∩ E)) = µ(En ∩ E) = µE (En ). n∈N n∈N Therefore µE is a measure. 11. Suppose that µ is continuous from below. Let {En }n∈N be a pairwise disjoint collection of measurable sets, and for each n ∈ N define Fn := ∪ni=1 En . Since µ is continuous from below and finitely additive, µ(∪i∈N Ei ) = µ(∪n∈N Fn ) = lim µ(Fn ) = lim n→∞ n→∞ n X i=1 µ(Ei ) = ∞ X µ(Ei ). i=1 This shows that µ is a measure. Conversely, if µ is a measure it is continuous from below by Theorem 1.8. Now suppose that µ(X) < ∞ and µ is continuous from above. If (En )n∈N is an increasing sequence of measurable sets, then (Enc )n∈N is a decreasing sequence of measurable sets, which implies that µ(∪n∈N En ) = µ(X) − µ((∪n∈N En )c ) = µ(X) − µ(∩n∈N Enc ) = µ(X) − lim µ(Enc ) = lim µ(En ). n→∞ n→∞ This shows that µ is continuous from below, so it is a measure by the previous paragraph. Conversely, if µ is a (finite) measure it is continuous from above by Theorem 1.8. 2 Real Analysis Chapter 1 Solutions Jonathan Conder 14. Suppose for a contradiction that there exists C ∈ (0, ∞) such that every measurable subset F ⊆ E satisfies µ(F ) ≤ C or µ(F ) = ∞. Set M := sup{µ(F ) | F ⊆ E is measurable and µ(F ) < ∞}, and note that 0 ≤ M ≤ C. For each n ∈ N there exists a measurable subset En ⊆ E such that M − n−1 ≤ µ(En ) < ∞. Set Fn := ∪ni=1 En for each n ∈ N P and define F := ∪n∈N Fn . Note that M − n−1 ≤ µ(En ) ≤ µ(F ) and also µ(Fn ) ≤ ni=1 µ(En ) < ∞ for all n ∈ N, so M ≤ µ(F ) = limn→∞ µ(Fn ) ≤ M. This shows that µ(F ) = M, so µ(E \ F ) = ∞. Since µ is semifinite, there exists a measurable subset A ⊆ E \ F such that 0 < µ(A) < ∞. This contradicts the definition of M, because A ∪ F ⊆ E but µ(F ) < µ(A) + µ(F ) = µ(A ∪ F ) < ∞. Therefore, for any C ∈ (0, ∞) there exists a measurable subset F ⊆ E such that C < µ(F ) < ∞. 17. Let A, B ⊆ X be disjoint µ∗ -measurable sets, and let E ⊆ X. Then (A ∪ B) ∩ Ac = B and hence µ∗ (E ∩ (A ∪ B)) = µ∗ (E ∩ (A ∪ B) ∩ A) + µ∗ (E ∩ (A ∪ B) ∩ Ac ) = µ∗ (E ∩ A) + µ∗ (E ∩ B). It follows by induction that Pn k=1 µ ∗ (E ∩ Ak ) = µ∗ (E ∩ (∪nk=1 Ak )) ≤ µ∗ (E ∩ (∪∞ k=1 Ak )) for all n ∈ N. Therefore ∞ X µ∗ (E ∩ Ak ) ≤ µ∗ (E ∩ (∪∞ k=1 Ak )), k=1 which implies that µ∗ (E ∩ (∪∞ k=1 Ak )) = P∞ k=1 µ ∗ (E ∩ Ak ) because µ∗ is subadditive. 18. (a) Let E ⊆ X and ε ∈ (0, ∞). If µ∗ (E) = ∞ then X ∈ A ⊆ Aσ , E ⊆ X and µ∗ (X) ≤ µ∗ (E) + ε. Otherwise ( ) X ∗ µ0 (An ) An ∈ A for all n ∈ N and E ⊆ ∪n∈N An , µ (E) = inf n∈N so there exists a sequence (An )n∈N in A such that E ⊆ ∪n∈N An and P then A ∈ Aσ , E ⊆ A and µ∗ (A) ≤ n∈N µ0 (An ) ≤ µ∗ (E) + ε. P n∈N µ0 (An ) ≤ µ∗ (E) + ε. If A := ∪n∈N An , (b) Let E ⊆ X such that µ∗ (E) < ∞. Suppose that E is µ∗ -measurable. For each n ∈ N there exists Bn ∈ Aσ such that E ⊆ Bn and µ∗ (Bn ) ≤ µ∗ (E) + n−1 . Define B := ∩n∈N Bn , so that B ∈ Aσδ and E ⊆ B. Since µ∗ (E) < ∞, it follows that µ∗ (B \ E) ≤ µ∗ (Bn ∩ E c ) = µ∗ (Bn ) − µ∗ (Bn ∩ E) = µ∗ (Bn ) − µ∗ (E) ≤ n−1 for all n ∈ N. This implies that µ∗ (B \ E) = 0. Conversely, suppose there exists B ∈ Aσδ such that E ⊆ B and µ∗ (B \ E) = 0. If F ⊆ X, then µ∗ (F ∩ E) + µ∗ (F ∩ E c ) ≤ µ∗ (F ∩ B) + µ∗ (F ∩ B c ) + µ∗ (F ∩ (B \ E)) ≤ µ∗ (F ∩ B) + µ∗ (F ∩ B c ) = µ∗ (F ) because B ∈ Aσδ is µ∗ -measurable. Clearly µ∗ (F ) ≤ µ∗ (F ∩ E) + µ∗ (F ∩ E c ), so E is µ∗ -measurable. 19. If E ⊆ X is µ∗ -measurable, then µ∗ (X) = µ∗ (X ∩ E) + µ∗ (X ∩ E c ) = µ∗ (E) + µ∗ (E c ) by definition, so µ∗ (E) = µ∗ (E). Conversely, let E ⊆ X and suppose that µ∗ (E) = µ∗ (E). By the previous exercise, for each n ∈ N there exist An , Bn ∈ Aσ such that E ⊆ An , E c ⊆ Bn , µ∗ (An ) ≤ µ∗ (E) + n−1 and µ∗ (Bn ) ≤ µ∗ (E c ) + n−1 . If A := ∩n∈N An , then µ∗ (A \ E) ≤ µ∗ (An ∩ Bn ) = µ∗ (An ) − µ∗ (An \ Bn ) = µ∗ (An ) − µ∗ (Bnc ) = µ∗ (An ) − (µ∗ (X) − µ∗ (Bn )) ≤ µ∗ (E) + n−1 − µ∗ (X) + µ∗ (E c ) + n−1 3 Real Analysis Chapter 1 Solutions Jonathan Conder = µ∗ (E) − µ∗ (E) + 2n−1 = 2n−1 for all n ∈ N, and hence µ∗ (A \ E) = 0. Moreover E ⊆ A and A ∈ Aσδ , so by the previous exercise E is µ∗ -measurable. 20. (a) If (An )n∈N is a sequence in M∗ such that E ⊆ ∪n∈N An , then X X µ(An ). µ∗ (E) ≤ µ∗ (∪n∈N An ) ≤ µ∗ (An ) = n∈N n∈N Therefore µ∗ (E) ≤ µ+ (E), by definition. If there exists A ∈ M∗ such that E ⊆ A and µ∗ (A) = µ∗ (E), then µ+ (E) ≤ µ(A) + ∞ X µ(∅) = µ∗ (A) = µ∗ (E), n=2 µ∗ (E) µ+ (E). µ∗ (E) so = Conversely, suppose that = µ+ (E). By exercise 18 part (a), for each n ∈ N there exists An ∈ M∗σ = M∗ such that E ⊆ An and µ+ (An ) ≤ µ+ (E) + n−1 . Define A := ∩n∈N An , so that A ∈ M∗ and E ⊆ A. Moreover µ+ (A) ≤ µ+ (An ) ≤ µ+ (E) + n−1 for all n ∈ N, so µ+ (A) ≤ µ+ (E). It follows that µ∗ (A) = µ+ (A) = µ+ (E) = µ∗ (E), because A ∈ M∗ and E ⊆ A. (b) Let µ0 be a premeasure on an algebra A ⊆ M∗ such that µ∗ is induced by µ0 , and let E ⊆ X. By exercise 18 part (a), for each n ∈ N there exists An ∈ Aσ ⊆ M∗ such that E ⊆ An and µ∗ (An ) ≤ µ∗ (E) + n−1 . Define A := ∩n∈N An , so that A ∈ M∗ and E ⊆ A. Moreover µ∗ (A) ≤ µ∗ (An ) ≤ µ∗ (E) + n−1 for all n ∈ N, so µ∗ (A) ≤ µ∗ (E). It follows that µ∗ (A) = µ∗ (E), because E ⊆ A. By the previous exercise, this implies that µ∗ (E) = µ+ (E). Therefore µ∗ = µ+ . (c) Define µ∗ : 2X → [0, ∞] by µ∗ (∅) = 0, µ∗ ({0}) = 2, µ∗ ({1}) = 2 and µ∗ (X) = 3. Clearly µ∗ (A) ≤ µ∗ (B) for all A ⊆ B ⊆ X. Moreover µ∗ is subadditive, because µ∗ (X) < µ∗ ({0}) + µ∗ ({1}). Therefore µ∗ is an outer measure. Note that {0} is not µ∗ -measurable, because µ∗ (X) = 3 < 2 + 2 = µ∗ (X ∩ {0}) + µ∗ (X ∩ {0}c ). This implies that P µ+ ({0}) = inf{ n∈N µ∗ (An ) | An ⊆ X for all n ∈ N and An = X for some n ∈ N} = µ∗ (X) = 3 6= 2 = µ∗ ({0}). 21. Let E ⊆ X be locally µ∗ -measurable, and let A ⊆ X. If µ∗ (A) = ∞, then clearly µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ∗ (A). Otherwise, by exercise 18 part (a), for each n ∈ N there exists a µ∗ -measurable set An ⊆ X such that A ⊆ An and µ∗ (An ) ≤ µ∗ (A) + n−1 . In particular, if n ∈ N then µ∗ (An ) < ∞, and hence An ∩ E and An ∩ E c = An ∩ (An ∩ E)c are µ∗ -measurable (because E is locally µ∗ -measurable). It follows that µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ∗ (An ∩ E) + µ∗ (An ∩ E c ) = µ(An ∩ E) + µ(An ∩ E c ) = µ(An ) = µ∗ (An ) ≤ µ∗ (A) + n−1 for all n ∈ N, so µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ∗ (A). Clearly µ∗ (A) ≤ µ∗ (A ∩ E) + µ∗ (A ∩ E c ) in either case, so E is µ∗ -measurable. This shows that µ is saturated. 22. (a) Clearly µ|M = µ, so by exercise 6 from section 1.3 it suffices to show that M∗ = M. To this end, let A ∈ M. Then A = E ∪ F for some E ∈ M and F ⊆ X such that F ⊆ B for some B ∈ M with µ(B) = 0. Clearly A ⊆ E ∪ B ∈ M = Mσδ and µ∗ ((E ∪ B) \ A) ≤ µ∗ (B) = 0. By exercise 18 part (c), it follows that A ∈ M∗ . Conversely, let A ∈ M∗ . By exercise 18 part (c) there exists B ∈ Mσδ = M such that A ⊆ B and µ∗ (B \ A) = 0. Since B \ A ∈ M∗ , exercise 18 part (b) implies that B \ A ⊆ E and µ∗ (E \ (B \ A)) = 0 for some E ∈ M. Note that B \ E ∈ M and A ∩ E ∈ M∗ . Moreover µ∗ (A ∩ E) ≤ µ∗ (E \ (B \ A)) = 0, so there exists F ∈ M such that A ∩ E ⊆ F and µ∗ (F \ (A ∩ E)) = 0, by exercise 18 part (b). It follows that µ(F ) = µ∗ (F ) ≤ µ∗ (F \ (A ∩ E)) + µ∗ (A ∩ E) = 0. 4 Real Analysis Chapter 1 Solutions Jonathan Conder Since A = (B \ E) ∪ (A ∩ E), it follows that A ∈ M. Therefore M∗ = M. e and µ e ∈ M, b be the completion of µ. Given A ∈ M∗ with µ(A) < ∞, the converse of the previous exercise (b) Let E implies that A ∈ M (using exercise 18 part (b) instead of part (c)). Hence A = E ∪ F for some E ∈ M and F ⊆ X such that F ⊆ B for some B ∈ M with µ(B) = 0. In particular µ b(A) = µ(E) = µ∗ (E) ≤ µ∗ (A) < ∞. e is locally µ e ∩ A) < ∞ the previous exercise implies that e ∩ A ∈ M, as E b-measurable. Since µ∗ (E Therefore E e ∩ A ∈ M∗ (again using exercise 18 part (b) instead of part (c)). This shows that E e is locally µ∗ -measurable, E e ∈ M∗ by exercise 21. so E e ∈ M∗ and A ∈ M be such that µ Conversely, let E b(A) < ∞. Then A = E ∪ F for some E ∈ M and F ⊆ X such that F ⊆ B for some B ∈ M with µ(B) = 0. It follows that e ∩ A) ≤ µ∗ (A) ≤ µ∗ (E) + µ∗ (F ) = µ(E) + 0 = µ µ∗ (E b(A) < ∞. e ∩ A ∈ M∗ and hence This implies that A ∈ M∗ (by the previous exercise using part (b) instead of part (c)), so E e e ∩ A ∈ M (again by the previous exercise). This shows that E e is locally µ e ∈ M. E b-measurable, so E e \ M, then µ e = M∗ . By exercise 6 from section 1.3, µ e e Therefore M b and µ agree on M. If E ∈ M b(E) = ∞ by e ∗ definition. Moreover, if E ∈ M = M and µ(E) < ∞, then E ∈ M by the previous exercise (using part (b) e \ M. e instead of part (c)). This implies that µ b and µ also agree on M 23. (a) Let E := {(a, b] ∩ Q | a, b ∈ R}. Clearly ∅ = (0, 0] ∩ Q ∈ E. If (a1 , b1 ] ∩ Q ∈ E and (a2 , b2 ] ∩ Q ∈ E then their intersection is (max{a1 , a2 }, min{b1 , b2 }] ∩ Q ∈ E. Moreover, the complement of (a, b] ∩ Q ∈ E is ((−∞, a] ∩ Q) ∪ ((b, ∞] ∩ Q), which is a disjoint union of elements of E provided that a ≤ b. If a > b then the complement of (a, b] ∩ Q is just (−∞, ∞] ∩ Q ∈ E. This shows that E is an elementary family of subsets of Q, so the collection of finite disjoint unions of members of E is an algebra. If (a1 , b1 ] ∩ Q ∈ E and (a2 , b2 ] ∩ Q ∈ E are not disjoint, then their union is (min{a1 , a2 }, max{b1 , b2 }] ∩ Q ∈ E. Therefore A is the collection of finite disjoint unions of members of E, so A is an algebra. (b) Let M be the σ-algebra generated by A. Since Q is countable and M is closed under countable unions, it suffices to show that {x} ∈ M for all x ∈ Q. Given x ∈ Q and n ∈ N, it is clear that (x − n1 , x] ∩ Q ∈ M. Therefore {x} = ∩n∈N ((x − n1 , x] ∩ Q) ∈ M, as required. (c) By definition µ0 (∅) = 0. Let (En )n∈N be a sequence of disjoint members of A whose union lies in A. If En = ∅ P for all n ∈ N, then ∪n∈N En = ∅ and hence µ0 (∪n∈N En ) = 0 = n∈N µ0 (En ). Otherwise ∪n∈N En 6= ∅, and P Em 6= ∅ for some m ∈ N, so µ0 (∪n∈N ) = ∞ and n∈N µ0 (En ) ≥ µ0 (Em ) = ∞. Therefore µ0 is a premeasure on A. Define µ1 , µ2 : 2Q → [0, ∞] by ∞, E 6= ∅ ∞, E contains 2−n m for some m ∈ Z and n ∈ N µ1 (E) = and µ1 (E) = 0, otherwise. 0, E=∅ Clearly µ1 |A = µ0 = µ2 |A , because every non-empty interval contains a rational of the form 2−n m for some m ∈ Z and n ∈ N. For the same reason that µ0 is a premeasure, µ1 is a measure. A very similar argument implies that µ2 is a measure. But µ1 6= µ2 because µ1 ({3−1 }) = ∞ whereas µ2 ({3−1 }) = 0. 24. (a) Since µ(A) + µ(Ac ∩ B) = µ(A ∪ B) = µ(B) + µ(B c ∩ A), by symmetry it suffices to show that µ(Ac ∩ B) = 0. To this end, note that E ⊆ A ∪ B c = (Ac ∩ B)c (if a member of E is not in B c , it is in B ∩ E = A ∩ E ⊆ A). So µ(X) = µ∗ (X) = µ∗ (E) ≤ µ∗ ((Ac ∩ B)c ) = µ((Ac ∩ B)c ) = µ(X) − µ(Ac ∩ B), and hence µ(Ac ∩ B) = 0 as required. 5 Real Analysis Chapter 1 Solutions Jonathan Conder (b) It is clear that ME is a σ-algebra on E and that ν(∅) = 0. Let {En }n∈N be a pairwise disjoint subset of ME . For each n ∈ N there exists An ∈ M such that En = An ∩ E. Set A := ∪i∈N ∪∞ j=i+1 (Ai ∩ Aj ) and define Bn := An \ A for each n ∈ N. It is easily checked that {Bn }n∈N is a pairwise disjoint subset of M and that En = Bn ∩ E for all n ∈ N. Therefore ν(∪n∈N En ) = ν(∪n∈N (Bn ∩ E)) = ν((∪n∈N Bn ) ∩ E) = µ(∪n∈N Bn ) = X µ(Bn ) = n∈N X ν(En ), n∈N which shows that ν is a measure on ME . 25. Let (Cn )n∈N be a sequence of compact intervals covering R, and fix n ∈ N. There exists a Gδ set Vn ∈ Mµ and a null set Nn ∈ Mµ such that E ∩ Cn = Vn \ Nn . Let (Vnk )k∈N be a sequence of open sets such that Vn = ∩k∈N Vnk . For each k ∈ N define an open set Unk := Vnk ∪ Cnc , and set U := ∩n,k∈N Unk . Then U is a Gδ set and U \ E ⊆ ∪n∈N Nn . Indeed, if x ∈ U \ E there exists n ∈ N such that x ∈ Cn , which implies that x ∈ Vnk for all k ∈ N and hence x ∈ Vn but x∈ / E ∩ Cn = Vn \ Nn . Moreover E ⊆ U, because E ⊆ Unk for all n, k ∈ N. It follows that E = U \ (U \ E), where U \ E is a null set. If n ∈ N, then E ∩ Cn = Hn ∪ Nn for some Fσ set Hn ∈ Mµ and some null set Nn ∈ Mµ . Clearly ∪n∈N Hn and ∪n∈N Nn are respectively Fσ and null sets. Moreover E = ∪n∈N (E ∩ Cn ) = (∪n∈N Hn ) ∪ (∪n∈N Nn ). 26. Let E ∈ Mµ and suppose that µ(E) < ∞. Given ε ∈ (0, ∞), there exists an open set U ∈ Mµ such that E ⊆ U and µ(U ) < µ(E) + 2ε . Let (Un )n∈N be a sequence of disjoint open intervals such that ∪n∈N Un = U. Then X µ(Un ) = µ(U ) < ∞, n∈N so there exists N ∈ N such that P∞ n=N +1 µ(Un ) < 2ε . Define A := ∪N n=1 Un . It follows that µ(E4A) ≤ µ(E \ A) + µ(A \ E) ≤ µ(E \ U ) + µ(U \ A) + µ(U \ E) = 0 + ∞ X µ(Un ) + µ(U ) − µ(E) < ε. n=N +1 28. Let a, b ∈ R. Since µF is continuous from above, µF ([a, b]) = µF (∩n∈N (a − n−1 , b]) = lim µF ((a − n−1 , b]) = lim (F (b) − F (a − n−1 )) = F (b) − F (a−). n→∞ n→∞ It follows that µF ({a}) = µF ([a, a]) = F (a) − F (a−), in which case µF ([a, b)) = µF ([a, b]) − µF ({b}) = F (b−) − F (a−) and µF ((a, b)) = µF ([a, b)) − µF ({a}) = F (b−) − F (a). 29. (a) Suppose that E ⊆ N but m(E) > 0. Define R := Q ∩ [0, 1), and for each r ∈ R set Er := E + r. Clearly each Er is measurable with m(Er ) = m(E), and ∪r∈R Er ⊆ [0, 2). Let r, s ∈ R and suppose that Er intersects Es . Then there exists t ∈ Er ∩ Es , so that t − r, t − s ∈ E ⊆ N. Since t − r = (t − s) + (s − r) and s − r ∈ Q, the definition of N implies that t − s = t − r. Therefore r = s, which shows that {Er }r∈R is pairwise disjoint. Hence ∞= X r∈R m(E) = X m(Er ) = m(∪r∈R Er ) ≤ m([0, 2)) = 2, r∈R which is a contradiction. Therefore m(E) = 0. 6 Real Analysis Chapter 1 Solutions Jonathan Conder (b) Suppose that m(E) > 0, but every subset of E is measurable. Since E = ∪n∈Z (E ∩ [n, n + 1)), there exists n ∈ Z such that m(E ∩ [n, n + 1)) > 0. Define F := (E ∩ [n, n + 1)) − n, so that F ⊆ [0, 1), m(F ) > 0 and every subset of F is measurable. Also define R := Q ∩ [−1, 1], and for each r ∈ R set Nr := N + r. It is clear that [0, 1) ⊆ ∪r∈R Nr , and hence F = ∪r∈R (F ∩ Nr ). If r ∈ R then F ∩ Nr ⊆ F and (F ∩ Nr ) − r ⊆ N, which implies that both subsets are measurable (by containment in F and translational invariance) and have measure zero (by P P the previous exercise and translational invariance). Therefore m(F ) ≤ r∈R m(F ∩ Nr ) = r∈R 0 = 0, which is a contradiction so not every subset of E is measurable. 30. Let E ∈ L with m(E) > 0, and suppose there exists α ∈ (0, 1) such that m(E ∩ I) ≤ αm(I) for all open intervals I. Without loss of generality m(E) < ∞ (since m is semifinite, we may replace E by a subset of finite positive measure). Define ε := m(E)(1 − α), so that ε > 0. Since E ∈ L and m is outer regular, there exists an open set U ⊆ R such that E ⊆ U and m(U ) < m(E) + ε. As U is open, there exists a pairwise disjoint collection {Ii }i∈N of open intervals such that U = ∪i∈N Ii , and these intervals are bounded because m(U ) < ∞. If if i ∈ N then m(Ii ) = m(Ii \ E) + m(E ∩ Ii ) ≤ m(Ii \ E) + αm(Ii ) and hence (1 − α)m(Ii ) ≤ m(Ii \ E). Since ∪i∈N (Ii \ E) = U \ E has measure m(U ) − m(E), it follows that (1 − α)m(U ) = (1 − α) X i∈N m(Ii ) = X (1 − α)m(Ii ) ≤ i∈N X m(Ii \ E) = m(∪i∈N (Ii \ E)) < ε = (1 − α)m(E) i∈N and hence m(U ) < m(E) ≤ m(U ), which is impossible. Thus, for each α ∈ (0, 1) there exists an open interval I such that m(E ∩ I) > αm(I). The same clearly holds for α ∈ (−∞, 0]. 31. Let E ∈ L with m(E) > 0, and set α := 34 . By the previous exercise there exists a open interval I ⊆ R with endpoints a, b ∈ R such that m(E ∩ I) > αm(I). Suppose there exists x ∈ (− 12 m(I), 12 m(I)) such that x ∈ / E − E. Then −x ∈ / E − E, and clearly 0 ∈ E − E, so we may assume x > 0. Moreover E ⊆ (E c + x) ∩ (E c − x). Note that (a, a + 2x] ⊆ I, because 2x < m(I) = b − a. Define n := max{k ∈ N | a + 2kx < b}, so that m(E ∩ (a, a + 2nx]) = ≤ n X k=1 n X (m(E ∩ (a + 2(k − 1)x, a + (2k − 1)x]) + m(E ∩ (a + (2k − 1)x, a + 2kx])) (m((E c − x) ∩ (a + 2(k − 1)x, a + (2k − 1)x]) + m((E c + x) ∩ (a + (2k − 1)x, a + 2kx])) k=1 ≤ ≤ n X k=1 n X (m(E c ∩ (a + (2k − 1)x, a + 2kx] − x) + m(E c ∩ (a + 2(k − 1)x, a + (2k − 1)x] + x)) (m(E c ∩ (a + (2k − 1)x, a + 2kx]) + m(E c ∩ (a + 2(k − 1)x, a + (2k − 1)x])) k=1 ≤ m(E c ∩ (a, a + 2nx]). This implies that m(E ∩ (a, a + 2nx]) ≤ 12 (m(E ∩ (a, a + 2nx]) + 21 m(E c ∩ (a, a + 2nx])) = 21 m((a, a + 2nx]) = nx. Note that 4nx ≥ m(I), since otherwise k := 2n satisfies a + 2kx < b (but k > n). It follows that m(E ∩ I) ≤ m(E ∩ (a, a + 2nx]) + m((a + 2nx, b)) ≤ m(I) − nx ≤ m(I) − 3m(I) m(I) = = αm(I). 4 4 This is a contradiction, so there does not exist x ∈ (− 12 m(I), 12 m(I)) such that x ∈ / E − E. 7 Real Analysis Chapter 1 Solutions Jonathan Conder 33. Choose a surjection q : N → Q ∩ [0, 1], and for each n ∈ N define In := (q(n) − 3−n , q(n) + 3−n ). For each n ∈ N, ∞ define Dn = In \ (∪∞ k=n+1 Ik ). If m, n ∈ N and m < n, then Dm ∩ Dn = ∅ because Dn ⊆ In ⊆ ∪k=m+1 Ik . Moreover 2·3 −n = m(In ) ≤ m(Dn ) + m(∪∞ k=n+1 Ik ) ∞ X ≤ m(Dn ) + k=n+1 m(Ik ) = m(Dn ) + ∞ X 2 · 3−k = m(Dn ) + 3−n , k=n+1 and hence m(Dn ) ≥ 3−n , for all n ∈ N. For each n ∈ N there exists a Borel set An ⊆ Dn such that 0 < m(An ) < 3−n , which can be found by intersecting Dn with an interval of the form [3−n−1 k, 3−n−1 (k + 1)] for some k ∈ Z (not all of the intersections can have zero measure). Define A := [0, 1] ∩ (∪n∈N An ), and let I be a subinterval of [0, 1] with midpoint c. There exists n ∈ N such that 4 · 3−n < m(I), and there are infinitely many rationals in (c − 3−n , c + 3−n ), so there exists k ∈ N with k ≥ n and q(k) ∈ (c − 3−n , c + 3−n ). It follows that Ik ⊆ (c − 2 · 3−n , c + 2 · 3n ) ⊆ I. Therefore Ak ⊆ A ∩ I (since Ak ⊆ Ik ⊆ [0, 1]), so m(A ∩ I) ≥ m(Ak ) > 0. Moreover m(A ∩ I) = m((∪n∈N An ) ∩ I) = m(Ak ) + m((∪n∈N\{k} An ) ∩ I) < m(Dk ) + m(I \ Dk ) = m(I), because m(Ak ) < m(Dk ) and ∪n∈N\{k} An ⊆ ∪n∈N\{k} Dn ⊆ Dkc . 8