Math 318 HW #7 Solutions 1. Exercise 20.12. Let C be the Cantor set. Let D ⊂ [0, 1] be a nowhere dense measurable set with m(D) > 0. Then there is a non-measurable set B ⊂ D. At each stage of the construction of C and of D, a certain finite number of open intervals of [0, 1] are deleted (put into [0, 1]\C or [0, 1]\D). Let g map the intervals put into [0, 1]\D at the nth stage linearly onto the intervals put into [0, 1]\C at the nth stage, for n = 1, 2, . . .. Thus g is monotone and defined at every element of [0, 1]\D, mapping onto [0, 1]\C. (a) Using the fact that g is increasing and that [0, 1]\D is dense in [0, 1], prove that for every x0 ∈ D, lim g(x) = sup{g(x) : x ∈ [0, 1]\D and x < x0 } x→x− 0 and lim g(x) = inf{g(x) : x ∈ [0, 1]\D and x > x0 }. x→x+ 0 Proof. Since [0, 1]\D is dense, we know that every x0 ∈ D is a limit point of [0, 1]\D, so limx→x− g(x) makes sense. Now, 0 {g(x) : x ∈ [0, 1]\D and x < x0 } is nonempty (unless x0 = 0, in which case limx→x− g(x) is vacuous) and, since g is 0 increasing, bounded above by g(x1 ) for any x1 > x0 such that x1 ∈ [0, 1]\D. Therefore, sup{g(x) : x ∈ [0, 1]\D and x < x0 } = M exists. But then, for any > 0, there exists x0 ∈ [0, 1]\D with x0 < x0 such that g(x0 ) > M − . Letting δ = x0 − x0 , we know that x0 < x < x0 implies that g(x0 ) ≤ g(x) ≤ M , and so g(x0 ) − M ≤ g(x) − M ≤ 0. Therefore, since g(x0 ) − M > −, the above implies that − < g(x) − M ≤ 0, so we can conclude that |x − x0 | < δ and x < x0 implies that |g(x) − M | < . Since the choice of > 0 was arbitrary, we conclude that lim g(x) = M = sup{g(x) : x ∈ [0, 1]\D and x < x0 }, x→x− 0 as desired. An equivalent argument shows that limx→x+ g(x) = inf{g(x) : x ∈ [0, 1]\D and x > 0 x0 }. (b) Use (a) to show that for x0 ∈ D, limx→x− g(x) ≤ limx→x+ g(x). 0 1 0 Proof. Since g is monotone increasing, every element of the set {g(x) : x ∈ [0, 1]\D and x > x0 } is an upper bound on the set {g(x) : x ∈ [0, 1]\D and x < x0 }. Therefore, by definition of the supremum, lim g(x) = sup{g(x) : x ∈ [0, 1]\D and x < x0 } x→x− 0 is less than or equal to every element of {g(x) : x ∈ [0, 1]\D and x > x0 }. In other words, limx→x− g(x) is a lower bound on the set {g(x) : x ∈ [0, 1]\D and x > x0 }, and 0 so, by the definition of the infimum, is less than or equal to lim g(x) ≤ inf{g(x) : x ∈ [0, 1]\D and x > x0 } = lim g(x), x→x− 0 x→x+ 0 as desired. (c) Show that if a = limx→x− g(x) < limx→x+ g(x) = b for some x0 ∈ D, then C would 0 0 contain the interval (a, b), contradicting the fact that C is nowhere dense. Therefore limx→x− g(x) = limx→x+ g(x) = limx→x0 g(x), and the function 0 0 ( g(x) f (x) = limx0 →x g(x0 ) if x ∈ [0, 1]\D if x ∈ D is continuous and monotone. Proof. Suppose y ∈ (a, b). Then g(x) = y implies that x ∈ [0, 1]\D and that both x ≥ x0 and x ≤ x0 . These three conditions cannot all be satisfied simultaneously (since x0 ∈ / [0, 1]\D), so it must not be the case that g(x) = y for any x ∈ [0, 1]\D. But then y is not in the image of g, which is [0, 1]\C, so y ∈ C. But the above argument was independent of y, so it holds for all y ∈ (a, b), so the whole interval (a, b) ⊂ C, which is clearly impossible since m((a, b)) = b−a > 0 = m(C). From the contradiction, then, we conclude that lim g(x) = lim g(x), x→x− 0 x→x+ 0 as desired. (d) Let B0 = f (B). Show that B0 is measurable but f −1 (B0 ) = B is not, even though f is a measurable function. Since f −1 (B0 ) is not measurable, it is not a Borel set. By exercise 16.15, B0 is not a Borel set, even though it is measurable. Proof. Since B0 ⊂ C and C has measure zero, we know that B0 is measurable with measure zero. Now, B is non-measurable by construction, but we do need to make sure that f −1 (B0 ) = B. Of course, this is obvious since f is monotone and, hence, one-to-one. 2 (e) Using the fact that χB = χB0 ◦ f , show that composition in the opposite order in Corollary 17.7 is false. Answer. We check that, for any x ∈ [0, 1], ( 1 if f (x) ∈ B0 (χB0 ◦ f )(x) = χB0 (f (x)) = = χf −1 (B0 ) (x) = χB (x). 0 else Now, f is certainly continuous and, since B0 is measurable, χB0 is a measurable function. On the other hand, B is not measurable, so χB is not a measurable function. Therefore, the composition χB0 ◦ f provides a counterexample to the statement “If f : A → R is continuous and g : f (A) → R is measurable, then g ◦ f is measurable”. 2. Exercise 20.16. (a) Prove or disprove: if f = g a.e., and g is continuous, then f is continuous a.e. Counterexample. Let f be the Dirichlet function on [0, 1], i.e., ( 1 if x ∈ Q f (x) = 0 else for all x ∈ [0, 1], and let g(x) = 0 for all x ∈ [0, 1]. Then g is obviously continuous, and {x ∈ [0, 1] : f (x) 6= g(x)} = Q ∩ [0, 1] has measure zero, so f = g a.e. However, as we saw last semester, f is nowhere-continuous, so it is definitely not continuous a.e. (b) Prove or disprove the converse of (a). Counterexample. The converse of (a) is also false. Suppose ( 0 if x < 0 f (x) = 1 else. Then f is continuous except at x = 0; certainly {0} has measure zero, so f is continuous a.e. Now, suppose there were a continuous function g such that f = g a.e., meaning that A = {x ∈ R : f (x) 6= g(x)} has measure zero. Then certainly there exists x0 < 0 such that x0 ∈ / A, meaning that g(x0 ) = f (x0 ) = 0. Likewise, there exists x1 > 0 such that x1 ∈ / A, so g(x1 ) = f (x1 ) = 1. Thus, the Intermediate Value Theorem implies that there exists c ∈ (x0 , x1 ) such that g(c) = 1/2. Therefore, g −1 ((0, 1)) is open (since g is continuous) and non-empty (since c is an element), so m(g −1 ((0, 1))) > 0. But it must be the case that g −1 ((0, 1)) ⊂ A, which has measure zero. From this contradiction, then, we conclude that there cannot be a continuous function g such that f = g a.e., so f provides a counterexample to the converse of (a). 3 3. We say that a sequence (fn ) defined on a measurable set A ⊆ R converges in measure to a function f : A → R if lim m{x ∈ A : |fn (x) − f (x)| ≥ δ} = 0 n→∞ for all δ > 0. (a) Show that if (fn ) converges to f almost everywhere on A, then it converges to f in measure (cf. Exercise 20.26). Proof. Let δ > 0. Since (fn ) → f almost everywhere, the set B := {x ∈ A : lim fn (x) 6= f (x)} n→∞ has measure zero. For each k = 1, 2, . . ., define Bk := ∞ [ {x ∈ A : |fn (x) − f (x)| ≥ δ}. n=k Then certainly B1 ⊃ B2 ⊃ B3 ⊃ · · · and so, by Corollary 10.10, ! ∞ \ ∗ Bk = lim m∗ (Bk ) m k→∞ k=1 (1) Now, for each x ∈ / B, there exists N ∈ N such that n ≥ N implies |fn (x) − f (x)| < δ, so ∞ \ Bk ⊆ B, k=1 meaning that T∞ K=1 Bk has measure zero. By (1), then, lim m∗ (Bk ) = 0. k→∞ But now {x ∈ A : |fn (x) − f (x)| ≥ δ} ⊂ Bn for each n, so lim m∗ {x ∈ A : |fn (x) − f (x)| ≥ δ} = 0. n→∞ Since fn and f are both measurable, the function |fn − f | is measurable, meaning that the set {x ∈ A : |fn (x) − f (x)| ≥ δ} is measurable by Proposition 17.2(2), so the above implies that lim m{x ∈ A : |fn (x) − f (x)| ≥ δ} = 0, n→∞ as desired. (b) Suppose (fn ) converges in measure to f on A. Show that (fn ) converges in measure to a function g if and only if f = g (a.e.). 4 Proof. (⇒) Suppose (fn ) converges in measure to both f and g. Let B := {x ∈ A : f (x) 6= g(x)}; our goal is to show that m(B) = 0. Let x0 ∈ B. Then |f (x0 ) − g(x0 )| > 0, so, if k ∈ N such that |f (x0 ) − g(x0 )| ≥ 1/k, we have 1 ≤ |f (x0 )−g(x0 )| = |f (x0 )−fn (x0 )+fn (x0 )−g(x0 )| ≤ |f (x0 )−fn (x0 )|+|fn (x0 )−g(x0 )|. k This implies that 1 1 x0 ∈ x ∈ A : |fn (x) − f (x)| ≥ or |fn (x) − g(x)| ≥ 2k 2k 1 1 = x ∈ A : |fn (x) − f (x)| ≥ ∪ x ∈ A : |fn (x) − g(x)| ≥ 2k 2k for all n ∈ N. Therefore, if we let Bk = {x ∈ A : |f (x) − g(x)| ≥ 1/k}, we have that 1 1 ∪ x ∈ A : |fn (x) − g(x)| ≥ Bk ⊂ x ∈ A : |fn (x) − f (x)| ≥ 2k 2k so, by monotonicity of the Lebesgue measure, 1 1 m(Bk ) ≤ m x ∈ A : |fn (x) − f (x)| ≥ ∪ x ∈ A : |fn (x) − g(x)| ≥ 2k 2k 1 1 = m x ∈ A : |fn (x) − f (x)| ≥ + m x ∈ A : |fn (x) − g(x)| ≥ . 2k 2k Therefore, by the Order Limit Theorem, 1 1 + lim m x ∈ A : |fn (x) − g(x)| ≥ m(Bk ) ≤ lim m x ∈ A : |fn (x) − f (x)| ≥ n→∞ n→∞ 2k 2k = 0, so we can conclude that m(Bk ) = 0 for all k. S But now B = ∞ k=1 Bk and B1 ⊂ B2 ⊂ · · · , so Corollary 10.9 implies that ! ∞ [ m(B) = m Bk = lim m(Bk ) = lim 0 = 0. k→∞ k=1 k→∞ Thus, we conclude that f = g a.e. (⇐) On the other hand, suppose f = g a.e. Then B = {x ∈ A : f (x) 6= g(x)} has measure zero. Let δ > 0. Then {x ∈ A : |fn (x)−g(x)| ≥ δ} ⊂ {x ∈ A : |fn (x)−f (x)| ≥ δ}∪{x ∈ B : |fn (x)−g(x)| ≥ δ}. 5 The latter is a subset of B and, hence, has measure zero, so this implies that m ({x ∈ A : |fn (x) − g(x)| ≥ δ}) ≤ m ({x ∈ A : |fn (x) − f (x)| ≥ δ}) for all n, so indeed lim m ({x ∈ A : |fn (x) − g(x)| ≥ δ}) = 0 n→∞ for all δ > 0, and so (fn ) converges in measure to g. 4. Exercise 20.36. Prove that if f is monotone on a bounded measurable set A, then f is measurable on A. Proof. Let c ∈ R. Consider the set {x ∈ A : c ≤ f (x)}. This is a subset of A, and hence is bounded below, so it has an infimum, m. Essentially by Problem 3 from last semester’s take-home final, limx→m+ f (x) = y0 exists and, by construction, y0 ≥ c. Hence, for all x ≥ m such that x ∈ A, we have that f (x) ≥ y0 ≥ c. On the other hand, since f is increasing, f (x) ≥ c implies that x ≥ m. Therefore, {x ∈ A : c ≤ f (x)} = [m, +∞) ∩ A, which is measurable since it is the intersection of two measurable sets. Therefore, using Proposition 17.2(2), we see that f is a measurable function, as desired. 6