Math 516 Professor Lieberman April 24, 2009 HOMEWORK #7 SOLUTIONS Chapter 11 29. Let A, B be a Hahn decomposition for ν, and let E be a measurable set. Then we have νE = ν(E ∩ A) + ν(E ∩ B). Because A is a positive set, we have ν(E ∩ A) ≥ 0, so νE ≥ ν(E ∩ B). But the definition of ν − says that ν − E = ν(E ∩ B), and hence νE ≥ ν − E. Similarly, B is a negative set, so ν(E ∩ B) ≤ 0, and ν + E = ν(E ∩ A), so νE ≤ ν + E. 33. (a) Since X is σ-finite, there is a sequence (Xi ) of disjoint measurable sets with µXi < ∞ for all i and X = ∪Xi . Then, for each i, the finite-measure version of the Radon-Nikodym theorem gives a nonnegative integrable function fi on Xi such that Z νE = fi dµ E for any measurable subset E of Xi . We now define f on X by f (x) = fi (x) if x ∈ Xi . If E is a measurable subset of X, we set Ei = E ∩ Xi , so Z νEi = fi dµ, Ei and E = ∪Ei . Since ν is a measure, it follows that Z X XZ XZ νE = νEi = fi dµ = f dµ = f dµ. Ei Ei E Hence f is the desired function. (b) Suppose f and g are two nonnegative functions on a σ-finite measure space (X, B, µ) such that Z Z f dµ = E g dµ E for all measurable sets E. Then there S is a sequence (Xi ) of disjoint measurable sets of finite measure with X = Xi . Fix numbers N > M > 0 and let E = {x : f (x) > N, g(x) < M } and Ei = E ∩ Xi . Then Z Z f dµ ≥ N µEi , g dµ ≤ M µEi . Ei Ei Since µEi < ∞ and N > M , it follows that µEi = 0 for all i, and hence µE = 0. We now note that [ {x : f (x) > g(x)} = {x : f (x) > N, g(x) < M }, N >M >0 are in Q 1 2 so µ({x : f (x) > g(x)} = 0. Similarly, µ({x : f (x) < g(x)} = 0, so f = g a.e. h i dν . First, suppose f is a nonnegative, 34. (a) For simplicity of notation, we set g = dµ P simple function: f = ai χEi . Then Z X X Z f dν = ai ν(Ei ) = ai g dµ Ei Z X X Z ai χEi g dµ = ai χEi g dµ = Z · ¸ dν = f dµ. dµ If f is a nonnegative, measurable function, then it is the limit of an increasing sequence (fn ) of nonnegative simple functions, hence Z Z f dν = lim fn dν by the Monotone Convergence Theorem. In addition, (fn g) is an increasing sequence of measurable functions which converges to f g, so another application of the Monotone Convergence Theorem says that Z Z lim fn g dµ = f g dµ. R R Since fn dν = fn g dµ, we conclude that Z Z f dν = f g dµ in this case, as well.h i i (b) Now we write gi = dν for i = 1, 2. Then, for any measurable set E, we have dµ Z Z (ν1 + ν2 )(E) = ν1 (E) + ν2 (E) = g1 dµ + g2 dµ E E Z = (g1 + g2 ) dµ. E h i Since the Radon-Nikodym derivative d(ν1dµ+ν2 ) is unique, it follows that · ¸ · ¸ · ¸ d(ν1 + ν2 ) dν1 dν2 = + . dµ dµ dµ (c) Now we set · ¸ · ¸ dν dµ f= , g= . dµ dλ Let E be a measurable set. From the definition of the Radon-Nikodym derivative, we have that Z ν(E) = f dµ, E 3 and part (a) implies that Z Z f dµ = f g dλ. E E Hence Z ν(E) = f g dλ E for any measurable set E. Again the uniqueness of the Radon-Nikodym derivative implies that · ¸ dν fg = . dλ (d) Since · ¸ dµ , 1= dµ we infer from part (c) that · ¸· ¸ dµ dν 1= . dν dµ and hence · ¸ · ¸−1 dν dµ = . dµ dν 39. First, we have ν ¿ µ because, if µ(E) = 0, then E = ∅, so ν(E) = 0. For any point x ∈ X, we have that ν({x}) = 0 while, for any function f , we have Z f dµ = f (x). {x} Hence any function f which satisfies Z ν(E) = f dµ E for all E ∈ B must be identically zero. But this choice for f only works if ν(E) = 0 for all E ∈ B. Since [0, 1] ∈ B and ν([0, 1]) = 1, it follows that there is no such function f . 48. First, if E ∈ B, then for every horizontal or vertical line L, L∩E or L\E is countable, so L \ Ẽ = L ∩ E or L ∩ Ẽ = L \ E must be countable. Hence Ẽ ∈ B. Next, if (Ei ) is a sequence in B, let L be a horizontal or vertical line and set E = ∪Ei . If L ∩ Ei is countable for all i, then L ∩ E is a countable union of countable sets so it is countable. On the other hand, if L \ Ei is countable for some i, then L \ E ⊂ L \ Ei , so L \ E is countable. Hence E ∈ B. Therefore B is a σ-algebra. To see that µ is a measure, we just have to check countable additivity. Let (Ei ) be a countable collection of disjoint subsets of X and set E = ∪Ei . Next, write Ki for the collection of horizontal or vertical lines L such that L \ Ei is countable, and write K for the collection of horizontal or vertical lines L such that L \ E is countable. I claim that K = ∪Ki and that Ki ∩ Kj = ∅ if i 6= j. Since L \ E = ∩(L \ Ei ), it 4 follows that, if L \ E is countable, then L \ Ei is countable for some i, so K ⊂ ∪Ki . Conversely, if L\Ei is countable for some i, then L\E ⊂ L\Ei , so L\E is countable. Hence ∪Ki ⊂ K, so K = ∪Ki . Next, let L ∈ Ki . Then L \ Ei is countable. If L \ Ej is countable (for this same line L), then L \ (Ei ∩ Ej ) = (L \ Ei ) ∪ (L \ Ej ) is also countable, so L ∩ (Ei ∩ Ej ) is non-empty. Since Ei ∩ Ej = ∅ for i 6= j, it follows that i = j in this case. Hence L ∈ Ki for some i implies that L ∈ / Kj for i 6= j. Hence, the number of elements in K is the sum of the number of elements in each Ki . Hence µ is a measure. A similar argument (but just using horizontal lines) shows that ν is a measure. Next, if f ∈ L1 (µ), then ¯ Z ¯Z Z ¯ ¯ ¯ f dν ¯ ≤ |f | dν ≤ |f | dµ, ¯ ¯ so F is a bounded linear functional on L1 (µ). R Now, suppose that1 there is a locally measurable function g such that F (f ) = f g dµ for all f ∈ L (µ). Let H be a horizontal line. Then, for any vertical line L, L \ H is uncountable. In addition, if H 0 is a horizontal line H 0 \ H is equal to H 0 and hence uncountable if H 0 6= H while H 0 \ H = ∅ and hence countable if H 0 = H. Therefore µH = 1. We now set Z = {hx, yi : g(x, y) = 0}. Then, for any horizontal line H, Z ∩ H is measurable, so either H ∩ Z or H \ Z must be countable. If H \ Z were countable, then we would have 1 = ν(H ∩ Z) because ν(H ∩ Z) ≤ µH = 1 and H is a horizontal line such that H \ (H ∩ Z) = H \ Z is countable. On the other hand, Z ν(H ∩ Z) = g dµ = 0 H∩Z because g = 0 on H ∩ Z. It follows that H ∩ Z is countable for any horizontal line H. For a vertical line V , we have µV = 1 just as before, so Z ∩ V is measurable and hence either V ∩ Z or V \ Z must be countable. Now 0 = ν(V \ Z) because, for any vertical line H \ (V \ Z) is uncountable, and Z ν(V \ Z) = g dµ V \Z which implies that mu(V \ Z) = 0 because g 6= 0 on V \ Z. Hence V \ Z is countable for any vertical line V . We now use a definition that was very convenient in Chapter 12. For each x ∈ A, we write Zx = {hx, bi : b ∈ B} and, for each y ∈ B, we write Z y = {ha, yi : a ∈ A}. Then we consider the two possibilities on the comparison of the sizes of A and B. If A has fewer elements than B, we observe that \ Z= Zy, y∈B Since V \ Z is countable for each vertical line and Z y = V ∩ Z for some vertical line V , it follows that Z y is non-empty for each y ∈ B and hence Z has no fewer elements 5 than B. On the other hand, Z= \ Zx , x∈A and Zx = Z ∩ H for the vertical line H = {hx, bi : b ∈ B} so Z has no more elements than A. But we can’t have the three conditions (i) A has fewer elements than B, (ii) Z has no fewer elements than B, (iii) Z has no more elements than A. Therefore, there is no function g in this case. If A has more elements than B, we set W = (A × B) \ Z, and define Wx and W y as before (with W in place of Z). A similar argument shows that W has no more elements than B but no fewer elements than A. Again, there is no function g in this case. Chapter 12 2. First, we let E1 and E2 be disjoint measurable sets, we let A be any set and we set B = A ∩ (E1 ∪ E2 ). Then B ∩ E1 = A ∩ E1 and B ∩ E2 = A \ E1 . Because E1 is measurable, we conclude that µ∗ (A ∩ (E1 ∪ E2 )) = µ∗ B = µ∗ (B ∩ E1 ) + µ∗ (B \ E2 ) = µ∗ (A ∩ E1 ) + µ∗ (A ∩ E2 ). Now let (Ei ) be a sequence of disjoint measurable sets, set E = n [ Gn = Ei . i=1 A simple induction argument shows that n X ∗ µ (A ∩ Gn ) = µ∗ (A ∩ Ei ). i=1 By monotonicity, we conclude that ∗ µ (A ∩ E) ≥ n X µ∗ (A ∩ Ei ) i=1 for any positive integer n and hence µ∗ (A ∩ E) ≥ X µ∗ (A ∩ Ei ). On the other hand, countable subadditivity implies that X µ∗ (A ∩ E) ≤ µ∗ (A ∩ Ei ). Combining these two inequalities gives X µ∗ (A ∩ E) = µ∗ (A ∩ Ei ). S Ei and define 6 4. (a) Let A ∈ A and suppose A can be written as a finite union in two ways: A = Sn Sk i=1 Ci and A = j=1 Dj with Ci and Dj finite disjoint collections of sets in C . S Then for each i, we have Ci = kj=1 (Ci ∩ Dj ) and Ci ∩ Dj ∈ C , so µCi = k X µ(Ci ∩ Dj ). j=1 Hence n X µCi = i=1 n X k X µ(Ci ∩ Dj ). i=1 j=1 Similarly, k X µDj = i=1 k X n X µ(Dj ∩ Ci ). j=1 i=1 Because the sums are all finite, we have n X k X µ(Ci ∩ Dj ) = i=1 j=1 and hence k X n X µ(Dj ∩ Ci ), j=1 i=1 n X µCi = i=1 k X µDj . i=1 (b) Let (Ai ) be a countable disjoint collection of sets in A and suppose A = ∪Ai is in A . Then S there is a finite disjoint collection of sets {B1 , . . . , Bn } in C such that A = ni=1 Bi . In addition, for each positive integer i, there is a finite collection of sets {B1i ,S . . . , Bk(i)i } in C such that Ai = ∪Bji . We now Pset Bijk = Bji ∩ Bk . Then Bk = i,j Bijk , so property (ii) implies that µBk ≤ i,j µBijk . From part (a), we know that X X Bijk µA = µBk ≤ i,j,k and hence µA ≤ X µAi . 8. (a) If µ+ E = ∞, then the inequality is automatically true, so we may assume that µ+ E is finite. Now let ε P > 0 be given, and suppose that E ⊂ ∪Ai , where each + Ai is in A and µ E ≥ µ̄Ai − ε. Now set A = ∪Ai . Then µ∗ Ai = µ̄Ai so subadditivity implies that X X µ∗ A ≤ µ∗ Ai = µ̄Ai . by monotonicity, we know that µ∗ E ≤ µ∗ A, and hence µ∗ E ≤ µ+ E + ε. Since this inequality is true for any ε > 0, it follows that µ∗ E ≤ µ+ E. 7 (b) Suppose first that E is a set such that µ+ E = µ∗ E.If µ∗ E = ∞, then X is a µ∗ -measurable set with E ⊂ X and µ∗ X = µ∗ E. If µ∗ E < ∞, then, for any (n) positive integer n, there is a sequence (Ai ) of µ∗ -measurable sets such that S (n) E ⊂ i Ai and X (n) 1 1 µ̄Ai ≤ µ+ E + = µ∗ E + . n n i S (n) P ∗ (n) Setting A(n) = i Ai , we see that T µ∗ A(n) ≤ µ Ai by subadditivity, so 1 ∗ (n) ∗ (n) µ A ≤ µ E + n . We now set A = A and use monotonicity to conclude that A is µ∗ -measurable with µ∗ A ≤ µ∗ E + n1 for any positive integer n and hence µ∗ A ≤ µ∗ E. Since E ⊂ A, it follows from monotonicity that µ∗ A ≥ µ∗ E so µ∗ A = µ∗ E. Now suppose that E is a set and that there is a µ∗ -measurable set A with E ⊂ A and µ∗ A = µ∗ E. Then µ+ E ≤ µ+ A = µ∗ A = µ∗ E. Using the inequality from part (a), we conclude that µ+ E = µ∗ E. (c) Suppose first that µ+ E = µ∗ E for all sets E. Then part (b) gives a set A ⊃ E such that µ∗ A = µ∗ E, so µ∗ A < µ∗ E + ε for any ε > 0. (In this case, we can use the same set A for any ε. Hence µ∗ is regular. Conversely, suppose µ∗ is regular and fix a set E. Since µ∗ is regular, for each positive integer n, there is a µ∗ -measurable set An such that E ⊂ An and µ∗ An ≤ µ∗ E + n1 . If we set A = ∩An , then E ⊂ A and µ∗ A ≤ µ∗ E + n1 for any n. Since µ∗ A ≥ µ∗ E, we conclude that µ∗ A = µ∗ E, so µ+ E = µ∗ E by part (b). (d) If µ∗ is regular, then parts (b) and (c) imply that µ+ E = µ∗ E for any set E. Hence µ∗ is induced by µ̄. Conversely, if µ∗ is induced by a measure on an algebra, then we must have that µ+ E = µ∗ E for all sets E. Hence µ∗ is regular by part (c). (e) Write the set X = {a, b} and define µ∗ by µ∗ ∅ = 0, µ∗ {a} = 1, µ∗ {b} = 2, and µ∗ X = 2.5. It’s easy to check that µ∗ is an outer measure and that {b} is measurable while {a} is not. In addition µ∗ is not regular because {a} is a subset E of X but there is no measurable subset A of X with E ⊂ A and µ∗ A < µ∗ E + 1. P 19. Fubini’s Theorem says that if i,j |xij | < ∞ (in the sense of problem 2.22), then the following sums are all finite: ∞ ∞ ∞ X ∞ ∞ X ∞ X X X X |xij |, |xij |, |xij |, |xij |. i=1 In addition, i=j ∞ X ∞ X j=1 i=1 j=1 i=1 xij = X i,j xij = i=1 j=1 ∞ X ∞ X xij . i=1 j=1 Tonelli’s Theorem says that if all the numbers xij are nonnegative, then ∞ X ∞ ∞ X ∞ X X X xij . xij = xij = j=1 i=1 i,j i=1 j=1 8 Parts (i), (i)0 , (ii), and (ii)0 of Tonelli’s Theorem are true for any nonnegative function xij because all functions are measurable in this case.