Math 501 Introduction to Real Analysis Instructor: Alex Roitershtein Iowa State University Department of Mathematics Summer 2015 Exam #1 Solutions This is a take-home examination. The exam includes 8 questions. The total mark is 100 points. Please show all the work, not only the answers. 1. [12 points] Fix any d ∈ N and let k · k denote the usual Euclidean norm in Rd . That is, v u d uX x2i for a vector x = (x1 , . . . , xd ) ∈ Rd . kxk = t i=1 Suppose that u, v ∈ Rd . Find w ∈ Rd and r > 0 such that kx − uk = 2kx − vk if and only if kx − wk = r. Hint: Prove that w = 1 3 4v − u and r = 2 3 v−u . First solution: This is a short solution based on a “completion to square” idea. Start with the equation kx − uk2 = 4kx − vk2 . (1) Observe that it is equivalent to kxk2 − 2x · u + kuk2 = 4kxk2 − 8x · v + 4kvk2 , and hence to kxk2 − 2x · 4v − u kuk2 − 4kvk2 = 3 3 Let w = 31 4v − u . Then the left-hand side is kxk2 − 2x · w. Add to both the sides kwk2 , to obtain kuk2 − 4kvk2 + kwk2 3 kuk2 − 4kvk2 1 4kvk2 − 8v · u + 4kuk2 = + k4v − uk2 = 3 9 9 4 = kv − uk2 = r2 , 9 kx − wk2 = kxk2 − 2x · w + kwk2 = where r = 2 3 v−u . 1 Remark. This solution is using the hint, in particular it enjoys the a-priori knowledge of w. To guess the value of w and r one can find two pints satisfying equation (1) on the straight line connecting u and v and then observe that w must be the center of the segment connecting these two points. The location of the two points is given by Equation (8) below. An easy way to find them is to set x = v + t(u − v) = u + (1 − t)(v − u), plug these expressions into, correspondingly, the right- and left-hand sides of equation (1), and solve the ressulting quadratic equation to find two possible values of t. Second solution: This solution is longer than the previous one. It exploits the same orthogonal decomposition x = v − t(v − u) + z⊥ as the solution of a similar problem in the sample, and in fact is mimicking the latter. Assume that kx − uk2 = 4kx − vk2 . (2) x = v − t(v − u) + z⊥ , (3) Write where (v − u) · z⊥ = 0 and t ∈ R is a suitable constant. Thus −t(v − u) and z⊥ are projection of the vector x − v into, respectively, the subspace spanned by v − u and its orthogonal complement. Notice that (3) is equivalent to x = u + (1 − t)(v − u) + z⊥ . (4) Thus, plugging the above expressions for x into (2) (respectively, (3) into the left-hand side and (4) into the right-hand side) and using the Pythagoras theorem, we obtain (1 − t)2 kv − uk2 + kz⊥ k2 = 4t2 kv − uk2 + 4kz⊥ k2 . It follows that (2) is equivalent to kz⊥ k2 = (1 + t)(1 − 3t) kv − uk2 . 3 (5) For a fixed value of the parameter t, the right-hand side of (5) is a constant independent of x. In general, in view of (3), for any constant a > 0, kz⊥ k = a is an equation of a sphere with center at v − t(v − u) in the (d − 1)-dimensional subspace orthogonal to the vector v − u. (6) Since the right-hand side expression in the equation (5) must be non-negative, we obtain that (1 + t)(1 − 3t) ≥ 0, and hence −1 ≤ t ≤ 1/3. 2 (7) Notice that kx − wk = r is an equation of the d-dimensional sphere with center at w and radius equal r. In particular, the values of w and r, if exist, are uniquely determined by (2). It follows from (6) and (7) that if such a sphere exists its center must correspond to the middle of the segment with the endpoints (corresponding to the endpoints of the interval in (7) describing the range of t): x1 = v − (−1)(v − u) and 1 x2 = v − (v − u). 3 (8) By the symmetry, the center w of the sphere kx − wk = r should be located in the mid-point of this segment, namely at v − t0 (v − u) with t0 = −1+1/3 = − 31 being the mid-point of the 2 interval (−1, 1/3). Thus 1 1 w=v− − · (v − u) = 4v − u . 3 3 It follows then from (5) that 1 2 r2 = kx − wk2 = t + ku − vk2 + kz⊥ k2 3 (1 + t)(1 − 3t) 4 1 2 ku − vk2 + kv − uk2 = kv − uk2 , = t+ 3 3 9 and hence r = 23 kv − uk. As we mentioned before, the values of w and r are uniquely determined by (2), and hence the problem is solved once one admissible pair (w, r) is found. 2. [14 points] Solve Exercise 35 in Chapter 1 of the textbook. Remark. Solutions of Exercises 2 and 3 are somewhat similar, both are based on a version of Cantor’s diagonal trick. However, the solution to the former is slightly more technically involved, while the solution to the latter relies on the main idea in a more direct and transparent way. It therefore might be of a benefit to students to first read the solution to Exercise 3. Solution: (a) Functions with the values in the set {0, 1} are commonly called binary functions. If f is a binary function and A = {s ∈ S : f (s) = 1}, then f is called the indicator of set A and is denoted by 1A . Let G : P → F be a function mapping the subsets of S into their indicators: G(A) = 1A . Notice that any binary function is an indicator of some (possibly empty) set. Furthermore, 1A = 1B clearly implies A = B. Thus G is a bijection from P onto F. The inverse function G−1 : F → P defined by G−1 (1A ) = A is the desired bijection from F onto P. The existence of the bijection shows that F and P are equivalent (have the same cardinality). 3 (b) First, notice that if S is empty or finite than card(P) = card(P) = 2card(S) . The first equality is the result in (a) while the second can be easily verified, for instance, by induction on the number of elements in S. In general, there is a trivial injection from S onto P defined by γ(s) = {s}, thus mapping s ∈ S into the singleton {s} ∈ P. To show that S is not equivalent to F (and hence is not equivalent to P) one can utilize a version of Cantor’s diagonal trick as follows. Assume that there is a bijection β : S → F. It will be convenient to denote β(s) as f (s) . Then f (s) : S → {0, 1} is a binary function for each s ∈ S, and, since β is a bijection, each element in F is f (s) for some s ∈ S. To see that this is in fact impossible, define a binary function h : S → {0, 1} by setting h(s) = not f (s) (s) = 1 − f (s) (s). (9) Since β is a bijection, we must have h = f (t) for some t ∈ S. However, if h = f (t) then h(t) = f (t) (t) contradicting (9). This shows that h 6= f (t) = β(t) for any t ∈ S, and hence β : S → F is actually not a bijection. Remark. Here is an instructive alternative solution to 2(b). It is based on a similar but different idea from the one hinted in the textbook. In particular, rather than compare S to F it directly compares S to P. Assume that there is a bijection α : S → P. Define E ∈ P as follows E = {s ∈ S : s 6∈ α(s)}. (10) Since α is a surjection, E = α(s) for some s ∈ S. However, • s ∈ E = α(s) directly contradicts (10). • s 6∈ α(s) implies, by virtue of (10), that s ∈ E = α(s). Either way we get a contradiction showing that, in fact, a bijection α : S → P doesn’t exist. 3. [12 points] Let E be the set of all real numbers x ∈ [0, 1] whose decimal expansion contains only the digits 4 and 7. Is E countable? Solution: One can utilize a version of Cantor’s diagonal trick as follows. Assume that E is countable and let f : N → E be a bijection from N to E (enumeration of the elements in E). Using decimal expansions write f (n) = 0.an,1 an,2 an,3 · · · = ∞ X k=1 4 10−k an,k , n ∈ N, where an,k ∈ {4, 7}, k ∈ N. Consider the element of E g = 0.b1 b2 b3 · · · = ∞ X 10−k bk , k=1 where bk ∈ {4, 7} are defined as follows: bk = not ak,k = 11 − ak,k , k ∈ N. (11) Since f is a bijection, we must have g = f (n) for some n ∈ N. However, if g = f (n) then bn = an,n contradicting (11). This shows that g 6= f (n) for any n ∈ N, and hence f is actually not a bijection. 4. [14 points] Fix any real number α > 1. Take x1 > xn+1 = √ α − x2n α + xn = xn + , 1 + xn 1 + xn α and define recursively n ∈ N. (a) Prove that x1 > x3 > x5 > . . . . (b) Prove that x2 < x4 < x6 < . . . . √ (c) Prove that limn→∞ xn = α. √ (d) Let εn = |xn − α|. Show that εn < cβ n for some constants c > 0 and β ∈ (0, 1). Solution: (a, b) We have: √ √ √ α + xn √ ( α − 1)( α − xn ) xn+1 − α = − α= 1 + xn 1 + xn (12) Hence, since α > 1, The sequence yn = xn − √ α alternates the signs (13) n α + α+x 2α + xn (1 + α) α + xn+1 1+xn = = , α+xn = 1 + xn+1 2xn + 1 + α 1 + 1+xn (14) Furthermore, xn+2 and hence xn+2 − xn = 2α + xn (1 + α) 2(α − x2n ) − xn = . 2xn + 1 + α 2xn + 1 + α The claims in (a) and (b) follow from (13) and (15) combined together. 5 (15) (c) It follows from the results in (a) and (b) that the sequences x2n and x2n+1 are monotone. Furthermore, (13) implies that √ x2n < α < x2n−1 , ∀ n ∈ N. Therefore the sequence are bounded and following limits exist: t1 = lim x2n n→∞ and t2 = lim x2n+1 . n→∞ Moreover, it follows from (14) that t1 and t2 are non-negative roots of the equation t= 2α + t(1 + α) . 2t + 1 + α 2 The last identity √ yields 2t + t(1 + α) = 2α + t(1 + α), which has a unique non-negative solution t = α. The proof of part (c) is complete. √ (d) Let εn = |xn − α|. It follows from (12) that √ ( α − 1)εn . εn+1 = 1 + xn Therefore, √ εn+1 α−1 lim =√ , n→∞ εn α+1 (16) which implies that the error εn decays asymptotically with an exponential rate. More precisely, pick any √α − 1 β∈ √ ,1 . α+1 It follows from (16) and the definition of the limit that there is N ∈ N such that n > N implies εn+1 < β < 1. εn Thus, for any m ∈ N, N +m−1 Y εk+1 εN +m = < β m. εN ε k k=N Using the substitution n = N + m we can rewrite this as εN n > N. εn < N β n , β Therefore, setting c = max 1≤k≤N nε o k βk , we obtain εn ≤ cβ n , ∀ n ∈ N. 6 5. [12 points] Fix any d ∈ N and let k · k denote the usual Euclidean norm in Rd . That is, v u d uX x2i for a vector x = (x1 , . . . , xd ) ∈ Rd . kxk = t i=1 We say that a sequence (xn )n∈N in Rd converges to x ∈ Rd and write limn→∞ xn = x if for any ε > 0 there exists Nε ∈ N such that n ∈ N and m, n > Nε ⇒ kxn − xk < ε. Call two converging sequences X = (xn )n∈N and Y = (yn )n∈N in Rn equivalent and write X ∼ Y if lim kxn − yn k = 0. n→∞ Let E be the set of all converging sequences in Rn . Show that ∼ is an equivalence relation in E. Solution: To show that ∼ is an equivalence relation we must verify the following three property of the relation: 1. X ∼ X for any X ∈ E (reflexivity) 2. X ∼ Y implies Y ∼ X for any X, Y ∈ E (reflexivity) 3. X ∼ Y and Y ∼ Z imply together X ∼ Z for any X, Y, Z ∈ E (transitivity) Take any X = (xn )n∈N ∈ E. Clearly, limn→∞ kxn − xn k = 0 and hence X ∼ X. Furthermore, if Y = (yn )n∈N ∈ E and limn→∞ kxn − yn k = 0 then limn→∞ kyn − xn k = 0, and hence X ∼ Y implies Y ∼ X. Finally, if Z = (zn )n∈Z ∈ E, the triangle inequality yields kxn − zn k ≤ kxn − yn k + kyn − zn k, and hence X ∼ Y and Y ∼ Z imply together X ∼ Z. 6. [12 points] (a) Let (sn )n∈R be a sequence of reals such that sn + sn−1 . 2 Show that sn is a Cauchy sequence and hence converges. sn+1 = (b) Let (sn )n∈R be a sequence of reals defined recursively by s2n−1 , 2 Find lim supn→∞ sn and lim inf n→∞ sn . s1 = 0, , s2n = s2n+1 = 1 + s2n . 2 Hint: Consider the sequences un = s2n and vn = s2n−1 separately. 7 Solution: (a) Let In be the closed interval with endpoints sn−1 and sn . If sn−1 = sn the interval is degenerate and consists of a single point. Let Ln be the length of the interval In , that is Ln = |sn − sn−1 |. Then Ln+1 = sn + sn−1 1 − sn = Ln . 2 2 Iterating, we obtain Ln+1 = 2−n L1 , and hence lim Ln = 0. (17) n→∞ Observe now that sn+1 is the mid-point of the interval In , and hence In+1 ⊂ In . In particular, sk ∈ In for any n ∈ N and k > n. By virtue of (17), this implies that sn is a Cauchy sequence and hence converges. (b) Let un = s2n and vn = s2n−1 . In this notation, the recursions in the statement of the problem can be rewritten as un = vn 2 and vn+1 = 1 + un . 2 It follows that vn+1 1 1 = + un 2 4 2 (18) 1 1 1 + un = + vn . 2 2 2 (19) un+1 = and vn+1 = To understand the structure of the sequences un and vn pick any constants α, β ∈ R and set un = α + xn , vn = β + yn . In the new notation, (18) and (19) become, respectively, α + xn+1 = 1 1 1 + α + xn 4 2 2 β + yn+1 = 1 1 1 + β + yn . 2 2 2 and Thus, if we could choose parameters α and β in such a way that α= 1 1 + α 4 2 (20) β= 1 1 + β, 2 2 (21) and 8 we would have xn+1 = 12 xn and yn+1 = 12 yn . The latter identities clearly imply lim xn = lim yn = 0. n→∞ Fortunately, (20) and (21) both have solutions, namely α = account (22), this yields lim un = α = n→∞ 1 2 (22) n→∞ and 1 2 and β = 1. Taking into lim vn = β = 1. n→∞ (23) It is not hard to verify, using the definition of the limit of a sequence and (23), that any converging subsequence of sn cannot include infinitely many elements from the sequence un and, at the same time, infinitely many elements from the sequence vn . This implies that any converging subsequence of sn converges to either 1 or 12 . Using the sequential definition of the lim sup and lim inf, this immediately yields lim sup sn = 1 and n→∞ 1 lim inf sn = . n→∞ 2 7. [12 points] Let (sn )n∈R be a sequence of reals and define n 1X tn = si . n i=1 (a) Prove that if limn→∞ sn = s then limn→∞ tn = s. (b) Give an example to show that tn can converge even though sn doesn’t. Solution: (a) Assume that limn→∞ sn = s. For ε > 0 and let Nε ∈ N be an integer such that n > Nε implies |sn − s| < ε. Then for n > Nε , we have n |tn − s| = n 1X 1X (si − s) ≤ |si − s| n i=1 n i=1 Nε n Nε 1 X 1X 1X |si − s| + ε≤ |si − s| + ε. ≤ n i=1 n i=N +1 n i=1 ε Taking n → ∞ while ε and Nε remain fixed in the both sides of the resulting inequality N |tn − s| ≤ ε 1X |si − s| + ε, n i=1 we obtain lim sup |tn − s| ≤ ε. n→∞ 9 Since ε > 0 is an arbitrary positive real, lim supn→∞ |tn − s| = 0 and hence lim |tn − s| = 0. n→∞ The latter result is equivalent to the desired assertion limn→∞ tn = s. (b) Consider sn = (−1)n . Clearly, the sequence sn diverges. However, tn = (−1)n+1 /n and hence limn→∞ tn = 0. 8. [12 points] (a) Use induction to show that if (x + 1/x) is integer then (xn + 1/xn ) is also integer for any n ∈ N. (b) Show that sup(A ∪ B) = max{sup A, sup B} for any sets A, B ⊂ R. Solution: (a) Let Sn = (xn + 1/xn ), n ∈ N. Verify that Sn · S1 = Sn+1 + Sn−1 , and hence Sn+1 = Sn · S1 − Sn−1 . The claim follows from this identity by a version of the induction argument. (b) The inequalities sup A ≤ sup(A∪B) and sup B ≤ sup(A∪B) follow from the inclusions A ⊂ (A ∪ B) and B ⊂ (A ∪ B). These two inequalities together imply that max{sup A, sup B} ≤ sup(A ∪ B). (24) On the other hand, sup A bounds from above any element of A while sup B bounds from above any element of B. It follows that max{sup A, sup B} bounds from above any element in A ∪ B. Hence sup(A ∪ B) = max{sup A, sup B}. The desired result follows from (24) and (25) combined together. 10 (25)