Problem 1 Part a Suppose that f : X → C is measurable and g : Y → C is measurable, and let h(x, y) = f (x)g(y). Consider a compact set E ⊂ C not containing zero (such sets generate the σ-algebra of measurable sets in C). We need to show that h−1 (E) is in the σ-algebra M ⊗ N . h−1 (E) is the set {(x, y) : h(x, y) ∈ E} which can be rewritten as {(x, y) : f (x)g(y) ∈ E} Or equivalently as {(x, y) : f (x) = p for some p ∈ E. } g(y) (This is where we use the assumption that 0 ∈ / E). This is the same as the statement {(x, y) : ∃a sequence zn of points in Q + iQ and a sequence pn of points in E 1 pn 1 such that |f (x) − zn | < and| − zn | < n g(y) n for all n. This equivalence follows from the compactness of E- a subsequence of p . We can rewrite this the pn ’s converges to a point p in E at which f (x) = g(y) last set as ∞ \ [ {(x, y) : |f (x) − z| ≤ n=1 z∈Q+iQ 1 p 1 and | − z| ≤ } n g(y) n where the sets inside are measurable by the definition of the product measure and the measurability of f and g. Part b Note that f and g must have σ-finite support because they are in L1 . h is supported on the product of the supports Sf of f and Sg of g, which is seen to be σ-finite: we can write it as a countable union of the rectangles whose sides are the sets approximating Sf and Sg . Furthermore, this set is measurable because h is measurable. Therefore, we can view f and g as functions on the σ-finite measure spaces (Sf , M|Sf , µ) and (Sg , N |Sg , ν), and we can view h as a function on the product of these measure spaces. Fubini’s theorem therefore applies to this product. 1 Problem 2 Note that f (x, y) = (1 − xy)−a is a nonnegative function for any a. Therefore, Tonelli’s theorem applies and these 3 integrals exist provided that any one of 1 them does. We can compute the integral in x first, giving (1−a)y (1−(1−y)−a+1 ) 1 if a 6= −1 and − y ln(1 − y) if a = 1. Note that this function is integrable when −a + 1 > −1; therefore, we need a < 2. Problem 3 Part a Note that the integrand changes sign whenever x is a multiple of π. Furtherx| on the interval [(n)π, (n + 1)π]. Note more, the integrand is larger than | sin nπ that | sin x| is invariant under translations by multiples of π, so the integral R (n+1)π | sin(x)| is independent of n and nonzero (in fact, it is equal to 2); (n)π P∞ 2 therefore, we can bound the Riemann integral below by the sum n=1 nπ , which diverges. This proves that the Lebesgue integral is infinite by, e.g., continuity of measure applied to the intervals (0, n]. Part b Notice that the value of the integral Z ∞ e−xy sin(x) dy 0 is precisely sin(x) x , for any nonzero x. Consider the integral Z b 0 Z ∞ e−xy sin x dy dx. 0 We can apply Fubini’s theorem to this integral because it is bounded above by the integral of sinx x from 0 to b, which is finite because sinx x has a finite limit as x → 0. The above integral is therefore equal to Z ∞ 0 Z b e−xy sin x dx dy. 0 The inner integral can be evaluated using integration by parts; the antiderivative −xy −by x+cos x) b−e−by cos b ; plugging in b and 0 for x gives 1−e y sin . comes out to be − e (yysin 2 +1 y 2 +1 Now, we desperately hope that we can interchange the limit in b with the outer integral in order to vastly simplify this expression. Note that for all b > 0 and all y > 0, we have the bound ye−by ≤ K for some constant K, so the numerator is trapped between two constants. Therefore, for any b, the function 2 3 is bounded above in absolute value by 1+y 2 , which is integrable. So no matter which sequence of b’s you pick, the dominated convergence theorem applies and we can therefore interchange the limit and integral. This gives Z ∞ 1 dy 2 y +1 0 which is equal to π 2 because the antiderivative is just arctan y. Problem 4 Part a First, I will show that ν + (E) is larger than the supremum of ν(F ) for all measurable F ⊂ E: to see this, perform a Hahn decomposition of E, and notice that ν(F ) = ν(F ∩ P ) + ν(F ∩ N ) ≤ ν(F ∩ P ) ≤ ν(P ) = ν + (E), where the last equality follows from the definition of ν + in the Jordan decomposition theorem. It’s easy to see that this supremum is attained and equal to ν + (E) just by taking F = P . Taking F = N gives the result for ν − . Part b Clearly |ν|(E) is no larger than the supremum: just perform the decomposition E1 = P . Conversely, if E1 , . . . , Ej is aP decomposition of E, we can Pand E2 = NP write j |ν(Ej )| = j |ν(Ej ∩ P ) + ν(Ej ∩ N )| ≤ j ν(Ej ∩ P ) − ν(Ej ∩ N ) = P P + − j |ν|(Ej ) = ν(E) as desired. j ν (Ej ) + ν (Ej ) = Problem 5 I’m assuming that µ1 , µ2 , ν1 , and ν2 are positive measures. I will first show that ν1 × ν2 is absolutely continuous with respect to µ1 × µ2 . Suppose that ν1 × ν2 (E) = M > 0 but µ1 × µ2 (E) = 0. Assume that E is 1 contained in a rectangle on which both Radon-Nikodym derivatives dµ dν1 and dµ2 dν2 are bounded by some number B (If not, just discard the part of E where this fails; if B is sufficiently large we are still left with a set of positive νmeasure because because the Radon-Nikodym derivative is finite a.e.). Then for any countable collection P of rectangles Rj covering E (arranged by decreasing area), we have that j ν1 × ν2 (Rj ) ≥ M ; however, we can select collections of rectangles (again contained in the product of the places P where the RadonNikodym derivatives satisfy the uniform bound) so that j µ1 × µ2 (Rj ) < ǫ for any ǫ > 0. Therefore, by the pigeonhole principle, at least one Rj , say, (1) (1) µ1 ×µ2 (Rj∗ ). Therefore, either µ1 (Rj ∗ )/ν1 (Rj ∗ ) or Rj ∗ satisfies ν1 ×ν2 (Rj ∗ ) ≥ M ǫq (2) (2) µ2 (Rj ∗ )/ν2 (Rj ∗ ) is at least M ǫ . This can be made arbitrarily large by taking ǫ sufficiently small, which contradicts the Lebesgue-Radon-Nikodym bound. 3 It remains to be seen that the Radon-Nikodym derivative of µ1 × µ2 with respect to ν1 × ν2 is the product Radon-Nikodym We clearly R of 1the R dµderivatives. ×µ2 dµ2 1 have for any rectangle that R dµ (x , x ) = (x ) (x ). Because 1 2 1 dν2 2 dν1 ×ν2 R dν1 these rectangles generate the product σ-algebra of X1 × X2 , this implies that the equality holds over general sets and not just rectangles. This implies that the functions are equal almost everywhere by Proposition 2.23b in Folland. 4