MA244 Analysis III Solutions. Sheet 2. NB. THESE ARE SKELETON SOLUTIONS, USE WISELY ! 0.1 Properties of regulated functions and their Integrals. 1. (Q.1) Pick any > 0. As f, g are regulated, there exist φ, ψ ∈ S[a, b]: ||f − φ||∞ < /2, ||g − ψ||∞ < /2. Notice that ψ + φ ∈ S[a, b] (Assignment 1). But ||f + g − φ − ψ||∞ ≤ ||f − φ||∞ + ||g − ψ||∞ < . Therefore f + g ∈ R[a, b]. To show that f g ∈ R[a, b], notice that regulated functions are bounded. (Assignment 1.) Therefore there exist M, N > 0: ||f ||∞ < M , ||g||∞ < N . Let φ, ψ ∈ S[a, b]: ||f − φ||∞ < δ/M , ||g − ψ||∞ < δ/N , Notice that φψ ∈ S[a, b]. (Assignment 1.) Then ||f g − φψ||∞ = ||f (g − ψ) + g(f − φ) + (f − φ)(ψ − g)||∞ ≤ ||f (g − ψ)||∞ + ||(f − φ)g||∞ + ||(f − φ)(ψ − g)||∞ ≤ 2δ + δ 2 /(M N ). For any > 0 the equation 2δ +δ 2 /(M N ) = has a positive solution and we are done. The last inequality above uses the following property of the sup-norm: ||ab||∞ = sup |a(x)||b(x)| ≤ sup |a(x)| sup |b(y)| = ||a||∞ ||b||∞ x x y 2. (Q.2) sin2 (t·) is continuous hence regulated on [a, b]. By Question 1, sin2 (t·)φ ∈ . By the R[a, b], so the question is well posed. Recall that sin2 (tx) = 1−cos(2tx) 2 Rb RL lemma, a cos(2tx)φ(x)dx → 0 as t → ∞. Therefore, Z lim t→∞ a b 1 sin (tx)φ(x)dx = 2 2 Z b φ(x)dx. a 3. (Q.3) Let h := g◦f . Then h(0) = 0 and h(x) = sign(x sin(1/x)) = sign(sin(1/x)) 1 1 for x > 0. So, h(x) = 1 for x ∈ ( 2π(n+1/2) , 2πn ) and h(x) = −1 for x ∈ 1 1 ( 2πn , 2π(n+1/2) ). Such a function is not regulated as for any step function φ in I, ||h − φ||∞ > minc∈R (|1 − c|, |1 + c|) ≥ 1. 4. (Q. 4) Let f be defined as follows: f (a + (b − a)/n) = 1/n, n ∈ N; f (x) = 0 if x∈ / a+(b−a)/N. Note that f ∈ R[a, b]: consider φk ∈ S[a, b] : φk (x) = f (x) for x ≥ a+(b−a)/k and φk (x) = 0 for x < a+(b−a)/k. Then ||f −φk ||∞ ≤ 1/k → 0 as k → ∞. Therefore, the sequence of step functions {φk } converges to f uniformly and f is regulated by definition. Of course, f itself is not a step function as the cardinality of its range is infinite. The integral: Z b Z b f = lim φk = 0. a k→∞ a 5. (Q.5) Let f (x0 ±) = limy→x0 ± f (y). According to the Theorem cited in the hint, if f ∈ R[a, b], then f (x±) exists for any x ∈ (a, b) and the corresponding one-sided limits exist at the boundary points. The first step is to show that the set of points S = {x ∈ [a, b] : |f (x+) − f (x−)| > } is finite. Notice that S is the set of all points of discontinuity of f with the jump size greater than . As f is regulated, there exists φ ∈ S[a, b] such that for any x ∈ [a, b], |f (x) − φ(x)| < /4. Taking one-sided limits of the above double inequality, we find after a simple manipulation: f (x+) − f (x−) − /2 ≤ φ(x+) − φ(x−) ≤ f (x+) − f (x−) + /2. Take x ∈ S . As |f (x+) − f (x−)| > , the above inequality implies that φ(x+) − φ(x−) 6= 0. But the set of discontinuities of a step function is finite. As we have just showed, that S is contained in the set of discontinuities of φ, we can conclude that |S | < ∞. Finally, let us consider the sequence of sets S1/n , n ∈ N . As it is easy to check, any point of discontinuity of f is in S1/n for some n. Note that S1 ⊂ S1/2 ⊂ S1/3 ⊂ . . . . The enumeration of the set of the set of all discontinuities can be done inductively: the points of S1 can be enumerated as S1 is finite. Assume that all points in S1/n have been enumerated by 1, 2, . . . Nn . As the set S1/(n+1) \ S1/n is finite, we can enumerate its points starting with Nn + 1. The result is the enumeration of all points in S1/(n+1) . As any discontinuity of f is contained in S1/n for some n, we have established a one-to-one correspondence between the set of all discontinuities of f and N, which means that this set is countable by definition. 0.2 Integration. 6. (Q.6) Let PN = N5 {0, 1, 2, . . . N } be a sequence of partitions of [0, 5]. Let φN : φN (0) = 0, φN |(5/N )(k−1,k] = ((5/N ) · k)2 , k = 1, 2, . . . , N be a sequence of step functions. Then ||f − φN ||∞ ≤ max1≤k≤N (((5/N ) · k)2 − ((2/N ) · (k − 1)2 )) ≤ 50/N → 0 as N → ∞. Therefore, φN ’s converge to f uniformly. By definition, Z 5 Z f = lim 0 N →∞ 0 5 3 X N N 5 X 5 2 φN = lim (5/N · k) == lim k2 N →∞ N N →∞ N k=1 k=1 PN 2 PN kx (N +1)x 2 But |x=0 = ∂x2 e ex −1−1 |x=0 = (1/3)N 3 + (1/2)N 2 + k=1 k = ∂x k=1 e (1/6)N . Substituting this into the previous formula and computing the limit, we get Z 5 53 f= . 3 0 Rb 1 Ra 1 dx + 2 log(x) dx. Each of the terms is differ7. (Q.7) By additivity, I = − 2 log(x) entiable by FTC2. As ∂a F (b) = 0, ∂a I = −1/ log(a), ∂b I = 1/ log(b). Ra 8. (Q.8) I(x) = −J(a(x)) + J(b(x)), where J(a) = 2 1/ log(t)dt. As J is differentiable on the domain of a, b and a, b are differentiable on R, the function I is differentiable on R. So, we can apply the chain rule: I 0 (x) = −J 0 (a(x))a0 (x) + J 0 (b(x))b0 (x) = − a0 (x) b0 (x) + . log a(x) log b(x) 9. (Q.9) (a) F 0 (x) = log10 (ex )ex = x10 ex p √ (b) G0 (x) = 2x 1 + (1 + x2 )4 − 2x 1 + x8 10 10 (c) H 0 (x) = 2x(e−x + e−x ) = 4xe−x (d) I 0 (x) = 1 √ 2 x cos(x2 ) + 1 x2 10 cos(1/x4 ) 10. (Q.10) For all the integrals below the integrands are continuous on the respective intervals of integration and we can apply the fundamental theorem of calculus for their evaluation. Re R1 (a) Substitution: t = log(x). Then 1 sin(log(x))/xdx = 0 sin(t)dt = 1 − cos(1). R log(3) R log(3) sinh(x) 0 2 (b) log(2) 1/ cosh (x)dx = log(2) cosh(x) dx = tanh(log(3))−tanh(log(2)) = 1/5. R 1/√2 1 R π/4 √ (c) Substitution: x = sin(t). Then 0 dx = sign(cos(t))dt = 2 0 1−x π/4. R e2 1 R2 (d) Substitution: t = log(x). Then e x log(x) dx = 1 1t dt = log(2). R1 100 (e) −1 x1001 e−x dx = 0 as the integrand is odd and the limits of integration are symmetric. 0.3 Improper integrals. Rn Rn Rn 11. (Q.11) Let bn = n−1 F , an = n−1 f . Note that |an | ≤ n−1 |f | ≤ bn . The fact R∞ that 0 F converges implies that the series ∞ X bn < ∞ n=1 P As |an | < bn , the series ∞ n=1 an convergesRabsolutely by the comparison test N for series (Analysis I). Therefore, limN →∞ 0 f exists and is finite. To finish R Rn the proof, we need to show that 0 f converges for any choice of sequence Rn : limn→∞ Rn = ∞. Let dxe be the smallest integer greater or equal than x. Then Z Z Z Z dRe R | f− 0 dRe f| ≤ 0 dRe |f | ≤ R R→∞ |f | ≤ bdRe −→ 0, dRe−1 as P∞ n=1 bn converges. Therefore, Z Z R f = lim lim R→∞ R→∞ 0 dRe f, 0 where the right hand side has already been shown to converge. R∞ R∞ 3 3 12. (Q.12) The integrals 0 e−x and 1 e−x dx diverge and converge simultane3 ously. But e−x ≤ e−x for any x ≥ 1. Therefore, Z ∞ Z ∞ −x3 e ≤ e−x dx = 1. 1 We conclude that 11. R∞ 0 −x3 e 1 dx converges by the comparison principle of Question 13. (Q.13) (1 − x4 ) = (1 − x2 )(1 + x2 ) = (1 − x)(1 + x)(1 + x2 ) ≥ (1 − x) for x ∈ [0, 1). Therefore, Z 1 Z 1 1 1 √ √ dx ≤ < ∞. 1−x 1 − x4 0 0 Therefore the elliptic integral converges by the comparison principle. R∞ RR 4 14. (Q.14) −R x3 dx = x4 |R −R = 0. However, −∞ xdx is divergent, as Z 0 lim R1 →∞ Z 3 x dx + lim −R1 R2 →∞ R2 x3 dx 0 does not exist. 15. (Q.15) R 100 R 100 1 R 100 1 < (a) 0 x1/3 +2x11/4 +x3 dx ≤ 0 x1/3 dx = lim↓0 x1/3 dx = lim↓0 23 x2/3 |100 ∞. So the integral converges by the comparison principle. R∞ (b) The integral 1 sin(x) converges. (Use the comparison principle with x2 R ∞ sin(x) R1 2 F (x) = 1/x .) Therefore, 0 sin(x) and dx diverge and converge 2 x x2 0 simultaneously. But Z 1 Z 1 sin(x) sin(x) 1 cos(x) lim = − lim | − dx ↓0 ↓0 x2 x x Z 1 sin(x) 1 1 = − lim | − cos(x) log(x) | − sin(x) log(x)dx , ↓0 x R1 which diverges logarithmically (verify that 0 log(x) sin(x)dx converges!). R∞ Therefore, 0 sin(x) dx diverges. x2 0.4 Uniform convergence. Pn−1 k 1 16. (Q.16) For any x ∈ (−1, 1), limn→∞ k=0 x = 1−x := f (x) pointwise. Let Pn−1 k 1−xn fn (x) = k=1 x = 1−x . Choose xn = 1 − 1/n ∈ (−1, 1). Then lim |fn (xn ) − n→∞ 1 xnn | = lim | | = lim n(1 − 1/n)n = +∞. n→∞ 1 − x n→∞ 1−x So choosing = 1 in the negation of the definition of the uniform convergence, we always find n and x ∈ (−1, 1) such that |f (x) − fn (x)| > 1. Therefore, P can k x does not converge uniformly in (−1, 1). On the other hand, for any k x ∈ [−R, R], where 0 ≤ R < 1, |fn (x) − f (x)| ≤ Rn → 0 for n → 0, 1−R which immediately implies the uniform convergence on [−R, R]. 17. (Q.17) Without restricting generality, we can consider symmetric intervals [−R, R]. For any x ∈ [−R, R], | n−1 k X x k=1 k! x −e | ≤ ∞ X Rk k=n ≤ ∞ Rk X Rk−n = k! k! k=n (n + 1)(n + 2) . . . (n + (k − n)) ∞ Rk X Rk Rk R = e → 0 as k → ∞, k! k=0 k! k! where the last step can be justified using Stirling’s formula. Therefore, the exponential sequence converges uniformly on [−R, R], hence on any finite subinterval of R. 18. (Q.18) ⇒. If fn → f uniformly on A, then for any > 0 there is N ∈ N: for any n > N and any x ∈ A, |f (x)−fn (x)| < /2. Therefore, supx∈A |fn (x)−f (x)| < and we get the definition of limn→∞ supx∈A |fn (x) − f (x)| = 0. ⇐. If supx∈A |fn (x) − f (x)| < , then for any x ∈ A |fn (x) − f (x)| < and the the uniform convergence follows immediately from limn→∞ supx∈A |fn (x) − f (x)| = 0. 19. (Q.19) Let Sn = x2 − x4 + x4 − x6 + . . . + x2n − x2n+2 = x2 − x2n+2 . Clearly, limn→∞ Sn (x) = x2 for |x| < 1 and limn→∞ Sn (x) = 0 for |x| = 1. Consider xn = 1 − 1/n ∈ (−1, 1). Then for any δ > 0, there is nδ ∈ N such that for any n > nδ x2n − Sn = x2n+2 = (1 − 1/n)2n+2 ≥ (1 − 1/n)n = en log(1−1/n) ≥ e−1−δ .. n Therefore, limn→∞ supx∈(−1,1) |Sn (x) − x2 | = 6 0 and the convergence cannot be uniform by Question 18. 20. (Q.20) Suppose the convergence Fn → F were not uniform. Negating the definition of the uniform convergence, we conclude that there must exist c > 0 and a strictly increasing sequence {pn }n≥1 ⊂ N such that sup (Fpn (x) − F (x)) > c, n = 1, 2, . . . x∈[a,b] As Fpn − F is continuous on the closed interval [a, b], it reaches its maximal value at some xpn ∈ [a, b]. The sequence {xpn }n≥1 is bounded. By the BoltzanoWeierstrass theorem it contains a subsequence {xpnk }k≥1 , which converges to a point x0 ∈ [a, b]. By construction, Fpnk (xpnk ) − F (xpnk ) > c, k = 1, 2, . . . As Fn (x) > Fm (x) for any x and m > n, we conclude that Fpn (xpnk ) − F (xpnk ) > c. for each n < nk . Taking the limit k → ∞ in the above inequality and using the continuity, we find that Fpn (x0 ) − Fp (x0 ) ≥ c > 0, n = 1, 2, 3, . . . , which contradicts the pointwise convergence Fn → F at x0 . Therefore, Fn must converge to F uniformly. November the 9th, 2015 Sergey Nazarenko and Oleg Zaboronski.