Homework 3 Solutions 1. (Exercise 7.9) Let {fn } be a sequence of continuous functions which converges uniformly to a function f on a set E . Suppose x ∈ E , {xn } ⊆ E , and xn → x. Claim. limn→∞ fn (xn ) = f (x) Proof. Let > 0. As fn → f uniformly, f is continuous. Hence, there is a δ > 0 such that whenever x, y ∈ E and |x − y| < δ , we have |f (x) − f (y)| < . As xn → x, there is an N such that whenever n ≥ N1 , we have |xn − x| < δ . Then, if n ≥ N1 , we have |f (xn ) − f (x)| < . Now, nd N2 such that whenever n ≥ N2 , we have supy∈E |fn (y) − f (y)| < . Then, if N = max{N1 , N2 } and n ≥ N , we have |fn (xn ) − f (x)| ≤ |fn (xn ) − f (xn )| + |f (xn ) − f (x)| < 2 so fn (xn ) → f (x). 2. (Exercise 7.10) Let (x) denote the fractional part of the real number x. Let f (x) ∞ X (nx) n2 n=1 = for x ∈ R. Claim. f is dened on R. If fn (x) = (nx) n2 , we have |fn (x)| ≤ M -test (theorem 7.10). Proof. Claim. 1 n2 , so this series converges uniformly on R by the Weierstrass Every point of Q is a discontinuity for f . Suppose z = pq ∈ Q be in lowest terms (so q > 0 and gcd(p, q) = 1). Let gn = nk=1 fn , so {gn } are the partial sums of our series. Consider gq . Note there is a δ > 0 such that I = (z − δ, z + δ) contains only z with denominator less than or equal to q . Hence, we see gq is continuous on I\{z} and gq has a jump disconP η 1 tinuity at z . Set η = gq (z−) − gq (z+) > 0. Let N be such that ∞ n=N n2 < 4 and such that N ≥ q . We note gN has similar properties to gq at z ; namely that there is a J = (z − δ2 , z + δ2 ) such that gN is continuous on J\{z} and such that gN has a jump discontinuity at z . Indeed, we see gN (z−) − gN (z+) ≥ gq (z−) − gN (z+). Hence, as supx∈R |gN (x)−f (x)| < η4 , we havem if f (z−) and f (z+) exist, f (z−)−f (z+) ≥ η −2( η4 ) = η2 > 0. If either of f (z−) or f (z+) do not exist, then f is not continuous at z , so in either case f is discontinuous at z . Proof. Claim. P Every point of R\Q is a point of continuity for f . 1 Let z ∈ R\Q. Let > 0 and let N be such that ∞ / Q, there is a δ1 > 0 n=N n2 < . As z ∈ such that for I = (z − δ1 , z + δ1 ), gN is continuous on I . Hence, we may nd a δ > 0 such that for x with |x − z| < δ , |gN (x) − gN (z)| < . Then, we have for |x − z| < δ , Proof. P |f (x) − f (z)| ≤ |f (x) − gN (x)| + |gN (x) − gN (z)| + |gN (z) − f (z)| < 3 so f is continuous at z . Claim. f ∈ R on every bounded interval Proof. For a given bounded interval and given n, gn has nitely many discontinuities, so gn is Riemann integrable on this interval. Hence, f is integrable on this interval as well as it is a uniform limit of the gn . 1 3. (Exercise 7.12) Let {fn } dened on (0, ∞) be Riemann integrable on [t, T ] whenever 0 < t < T < ∞. |fn | ≤ g , fn → f uniformly on every compact subset of (0, ∞), and ˆ ∞ g(x)dx < ∞ 0 Claim. ˆ n→∞ ˆ ∞ ∞ f (x)dx fn (x)dx = lim 0 0 ´∞ ´T We note the hypotheses give g ≥ 0. Hence, 0 g(x)dx makes sense as either limT →∞ limt→0 t g(x)dx ´T or limt→0 limT →∞ t g(x)dx. Indeed, we rst show these two limits exist (as possibly ∞) and are equal (that is, equal if nite or, if one is innite, then so is the other). Proof. ´T First, for xed T , we see t g(x)dx increases as t → 0. Hence, we may evaluate this limit in [0, ∞], say as LT . We note that if t < T < T 0 , then ˆ ˆ T0 ˆ T g(x)dx = T0 g(x)dx + t g(x)dx t T ´ T0 where T g(x)dx is constant in t (and nite), so either LT is nite for every T , or LT is innite for every T . Moreover, this shows that if T < T 0 , then LT ≤ LT 0 . Thus, we may evaluate the limit limT →∞ LT ´T in [0, ∞], as this is an increasing sequence, say as L. Similarly, we set limT →∞ t g(x)dx = Mt and M = limt→0 Mt , where the justication for taking this limit is similar. We claim M = L, where M, L ∈ [0, ∞]. First suppose L = ∞. By the above, we either have L = ∞ for all T or LT nite for all T with ´T T limT →∞ LT = ∞ . Suppose LT = ∞ for all T , so limt→0 t g(x)dx = ∞. We see ˆ ˆ T T →∞ and so if limt→0 T g(x)dx ≥ Mt = lim t g(x)dx t ´T g(x)dx = ∞, we have limt→∞ Mt = ∞, so M = ∞. Now, suppose LT < ∞ for all T but t limT →∞ LT = ∞. For N ∈ R, nd S such that whenever T > S , we LT > N . Hence, we have for T > S , ˆ ˆ U t→0 U →∞ t→0 T g(x)dx ≥ lim M = lim Mt = lim lim g(x)dx = LT > N t→0 t t so M = ∞. The analysis for M = ∞ =⇒ L = ∞ is similar and will be omitted here. Thus, M is innite if and only if L is innte. Suppose both are nite. We now show in this case that M = L. Let > 0. Find S such that if T > S , we have |LT − L| < . Likewise, nd s such that if t < s, we have |Mt − M | < . We have, for such t, T , |L − M | ≤ |L − LT | + |LT − Mt | + |Mt − M | < 2 + |LT − Mt | ´T ´R ´T Now, LT − Mt = limr→0 r g(x)dx − limR→∞ t g(x)dx. We look at LT − r g(x)dx. By denition, there ´T is a δT such that when r < δT we have |LT − r g(x)dx| < . We note for R0 > R we have ˆ ˆ R0 r ˆ R0 g(x)dx = g(x)dx + R g(x)dx r 2 R and so LR0 = LR + ´ R0 R g(x)dx. Hence, if |LR − ˆ ˆ R0 |LR0 − r g(x)dx| < , then ˆ R0 g(x)dx| = ||L + r ´R ˆ R g(x)dx − R0 g(x)dx − + R r g(x)dx| < R That is. we may take δT independent of T for T large (ie nd δS , and this will work for all T > S ). All ´T together this gives a δ such that if r < δ and T > S , then |LT − r g(x)dx| < . Similarly, we may nd an N such that if R > N and t < s, then |Mt − t < min(s, δ) and T > max(S, N ), we have ´R t g(x)dx| < . Hence, if |L − M | ≤ 2 + |LT − Mt | and ˆ |LT − Mt | ≤ |LT − ˆ T T g(x)dx| + | t g(x)dx − Mt | < 2 t so |L − M | < 4. This shows L = M . ´∞ ´∞ Now, we claim that 0 fn (x)dx and 0 f (x)dx make sense as well. Indeed, note |f | ≤ g and, for such ´T ´∞ a function, given > 0´ we may ´ nd t,´T such that | t f (x)dx − t0 f (x)dx| < (use the bounds above for g and the fact that | f | ≤ |f | ≤ g for various intervals. For such a t, T we know fn converges to f uniformly on [t, T ], and so there is an N such that if n ≥ N , we have ˆ ˆ T | T fn (x)dx − f (x)dx| < t t Hence, if n ≥ N , we have ˆ | ˆ ∞ fn (x)dx − 0 ˆ ∞ f (x)dx| ≤ ˆ T fn (x)dx − 0 0 < ˆ ∞ | t ˆ T fn (x)dx| + | fn (x)dx − t ˆ T f (x)dx| + | t ˆ T f (x)dx − t 3 which proves the claim. 4. (Exercise 7.14) Let f be a continuous real function on R such that 0 ≤ f ≤ 1, f (t + 2) = f (t) for P∞ t ∈ R, P f = 0 on [0, 31 ], and f = 1 on [ 32 , 1]. Let Φ(t) = (x(t), y(t)) where x(t) = n=1 2−n f (32n−1 t) and ∞ y(t) = n=1 2−n f (32n t). Claim. Φ is continuous and maps [0, 1] onto [0, 1]2 . Proof. As both x, y are uniformly convergent sequences of continuous functions (eg by the Weierstrass P∞ P∞ M -test), Φ is continuous. To show onto, let x0 = n=1 2−n a2n−1 and y0 = n=1 2−n a2n where ak ∈ {0, 1} for each k, so we've written a base-2 expansion of an arbitrary (x0 , y0 ) in [0, 1]2 (that every point has P∞ point −n−1 (2an ). We claim Φ(t0 ) = (x0 , y0 ). To such an expansion was covered in discussion). Set t0 = n=1 3 see this, we show f (3k t0 ) = ak . If we can show this, then x(t0 ) is ∞ X 2−n f (32n−1 t0 ) = n=1 ∞ X 2−n a2n−1 = x0 n=1 and likewise y(t0 ) = y0 . Write 3 k t0 = k−1 X 3k−n−1 (2an ) + 3−1 (2ak ) + n=0 ∞ X n=k+1 3 3k−n−1 (2an ) ∞ f (x)dx| 0 Now, P k−n−1 (2an ) is an integer multiple of 2 as n ≤ k − 1 in all summands, n=0 3 ∞ k−n−1 3 (2a )) from the property that f (t + 2) = f (t) for all t. We note n n=k+1 Pk−1 | ∞ X 3k−n−1 (2an )| ≤ n=k+1 ∞ X 3k−n−1 2 = 2 ∞ X 3−n−2 = n=0 n=k+1 and 3−1 (2ak ) is either 0 or 32 . Hence, 3−1 (2ak ) + on ak . Indeed, we see f (3k t0 ) = ak , as claimed. P∞ n=k+1 2 1 91− 1 3 so f (3k t0 ) = f (3−1 (2ak ) + = 23 1 = 92 3 3k−n−1 (2an ) lies in either [0, 31 ] or [ 23 , 1], depending 5. Let {fn } be a sequence of monotonically increasing function from [0, 1] to [0, 1]. Suppose fn converge pointwise on [0, 1] to f and that f is continuous. Prove fn → f uniformly. Proof. Let > 0. As [0, 1] is compact, f is uniformly continuous, so there is a δ > 0 such that if k for 0 ≤ k ≤ M . As |x − y| < δ , then |f (x) − f (y)| < . Let M > 1δ and consider the points xk = M fn → f , we can nd an N such that if n ≥ N , then |f (xk ) − fn (xk )| < for every k ∈ {0, . . . , M }. Now, let x ∈ [0, 1] be arbitrary and nd k such that xk ≤ x ≤ xk+1 . Suppose n ≥ N . Then, as fn is increasing, we have fn (xk ) ≤ fn (x) ≤ fn (xk+1 ). Now, |fn (xk ) − f (xk )| < and likewise for xk+1 , so f (xk ) − < fn (x) < f (xk+1 ) + . As each fn is increasing, we see f is also increasing (for x < y we have fn (y) − fn (x) ≥ 0 for all n so f (y) − f (x) ≥ 0), so f (xk ) ≤ f (x) ≤ f (xk+1 ). Thus, |fn (x) − f (x)| < f (xk+1 ) + − (f (xk ) − ) = f (xk+1 ) − f (xk ) + 2. From our choice of xk with uniform continuity, we have f (xk+1 ) − f (xk ) < , so |fn (x) − f (x)| < 3. As this holds independent of x, we have uniform convergence. 6. Let f : R → R be continuous and set fn (x) = Claim. fn 1 n Pn−1 k=0 f (x + nk ). converges uniformly on every bounded interval. ´ x+1 Let g(x) = x f (t)dt. Let I = [a, b] with a < b. We show fn converges uniformly to g on I . Let > 0. As f is uniformly continuous on [a, b + 1], we may nd a δ > 0 such that for x, y ∈ [a, b + 1], if |x − y| < δ , then |f (x) − f (y)| < . Let N > 1δ . Suppose n ≥ N , so n1 < δ . Then, for x ∈ [a, b], we have Proof. ˆ x+1 n−1 k 1X f (x + ) − f (t)dt| n n x k=0 n−1 X ˆ x+ k+1 n k (f (x + ) − f (t))dt| = | k n k=0 x+ n k+1 ˆ n−1 X x+ n k ≤ |f (x + ) − f (t)|dt k n k=0 x+ n n−1 X ˆ x+ k+1 n < dt |fn (x) − g(x)| = | k=0 = n = k x+ n n so fn converges uniformly to g on [a, b + 1]. 4