(976 supp. MATH Solutions Manual to Walter Rudin’s Principles of Mathematical Analysis Roger Cooke, University of Vermont| i | | | | | | Chapter 8 Some Special Functions Exercise 8.1 Define ie eu" (e £0), ron {e G23 Prove that f has derivatives of all orders at z = 0 and that f(")(0) = 0 for ae Solution. We have lim, tke-WV2" — 0 for all k = 0,+1,42,... by LHospital’s tule. It is easily shown by induction that there is @ polynomial p, such that f(x) = pa(t)e“/™" for #0. Assuming (by induction) that f)(0) = 0, we then have f+1)(9) = lim ga(t de Ve" = 0, where ga(2) = 2pn (2) eno Exercise 8.2 Let a;; be the number in the ith row and jth column of the array a 0 0 3 7-10 «0 i 2-10 80403 -1 so that Prove that 199130 CHAPTER 8 SOME SPECIAL FUNCTIONS Solution. ‘This is a routine computation: ae LY = YL] = Y[-1+0-29) while Lda = Exercise 8.3 Prove that LM LL 77 gt if a4; > 0 for all é and j (the case too = +00 may occur) Solution. In fact the only case that we need to consider is the case when one of the two sums is infinite. If either sum is finite, we merely invoke Theorem 8.3, which explicitly states that the two sums are equal, Hence if either sum is infinite, then both are. Exercise 8.4 Prove the following limit relations: w-1 (a) kim, =logb (b>0). = log(1 +2) z (0) lim ead nym @ tn (145) =¢ Solution. (a) Consider the function f(z) = 67 = e7!8*, The limit we are considering is /“(0). By the chain rule f@)= 108 log b,131 Now take x = 0. (0) Let y= log(1 +2), so that «=e — 1. It is easy to justify the relation ee eee since lim yoy (c) Consider the function (1+ 2)/? = e22 By part (b) lim(1 +2)? = eae. (d) As above, we have +2)" = [( < limit of the expression inside the brackets is e. 1/(z/n))* | , and by part (¢) the Exercise 8.5 Find the following limits (a) timmy op S=o42** (8) lima oo Gq {nV ~ 1) (@) lim, +o SS2S Solution, (a) This limit is f/(0), where f(x) = (1+:)"/* (by part (c) of the previous problem). Now for « # 0, we have (142) log(1 +2) —2 (+1) a) = (042) Since we know that the limit of the first factor is e, we need only consider the limit inside the brackets. Since (1+2)log(1+2) = (e- we can cancel 2? from the numerator and denominator of the expression in brackets, and we see that the limit of this expression is }. Hence the limit of J'(z) as z + O exists and equals §. It then follows from the mean-value theorem that f’(0) equals this limit (see the corollary to Theorem 5.12). (0) Write this expression as eM 1 een Since 982 tends to 0 as n — oo, this fraction tends to the derivative of e* at 0, Le., it tends to 1.132 CHAPTER 8. SOME SPECIAL FUNCTIONS (c) Write this expression as sinz ~ rcos2 © c0sz(1 — cosa)” We can then use either Maclaurin series or L’Hospital’s rule to prove that the limit is 2. (d) Write this expression as ( sing — rcost sin) cos and again either by Maclaurin series or L’Hospital’s rule the limit is 3. Exercise 8.6 Suppose f(c)f(y) = f(x + y) for all real z and y. (a) Assuming that f is differentiable and not zero, prove that F(z) where c is a constant. (0) Prove the same thing, assuming only that f is continuous. Solution. (a) Since f is not 0, it follows that f(0) = 1 (take z = y = 0 in the basic relation that defines f). It then follows that f’(z) = f(z) "(0), and hence that the function g(x) = e~*f") f(x) satisfies /(0) = 0 for all x. Therefore g(a) = 9(0) = f(0) = 1 for all z, ie., f(«) =e, where c= f'(0). (b) The relation f(x) f(y) = f(a + y) shows that either f(z) is always zero, or it is never zero. In the latter case, since f is continuous, it cannot change sign, and therefore (since f(0) = 1) it is always positive. Let g(z) = log(f(c)]. Then g(x +y) = g(x) +9(y), and g is continuous. It suffices then to show that (2) = cx for some constant c= g(1). To this end, we note that the additive property of g implies that g(0) = 0, 9(—) = —9(cr), and (by an easy induction) g(nz) = ng(z) for all integers n = 0,+1,+2,.... Consider the set of x such that g(x) = 9(1)x. Obviously 0 and 1 belong to this set. If a belongs to this set, so does na for any n, since g(na) = ng(a) = ng(1)a = 9(1)(na). Finally, if a belongs to this set, so does #, n = 1,2,..., since g(a) = g(n®) = ng(2). That is, g(2) = 29(a) = 19(1)a = (1). It now follows that r belongs to this set for all rational numbers r, that is, the two continuous functions g(z) and g(1)x have the same values at all rational numbers r. Since the rational numbers are dense, and the set of points at which two continuous functions are equal is a closed set, it follows that g(x) = 9(1)z for all z. Exercise 8.7 If 0 <2 < §, prove that133 Solution. ‘To show the left-hand inequality, consider the function f(z) = sinx ~ 22 on the interval 0 < x < %. We have f(0) = f($) = 0. Since f”"(z) = “sine < 0, the function f’(z) is strictly decreasing on this interval. Therefore it has at most one zero on this interval; by Rolle’s theorem, it has exactly one zero, Since j"(x) <0 at that point, the function f(z) has a maximum at that point. Therefore f(x) > 0 for0 (0) = 0 for all x > 0 (the restriction x < ¥ is superfluous in this case). Exercise 8.8 For n=0,1,2,..., and 2 real, prove that, |sinne| < n|sinz|. Note that this inequality may be false for other values of n. For instance, fale lene [sin gal > gisinsl Solution. The inequality is obvious if n = 0 or n = 1. Then by induction we have sinna| |sin((n — 1)z +2)| = |sin((n—1)z) cos + cos((n — 1)z) sin z| |sin((n —1)2)| + |sina| < < (n-DJsing| {sin | = |r] | sin | A stronger remark can be made: If ¢ is not an integer, then |sincr| > |el|sin |. Hence this inequality fails for x = unless c is an integer. Exercise 8.9 (a) Put sy =1+ (3) +---+(1/N). Prove that fim, (sar — log.) exists. (The limit, often denoted by 7, is called Euler's constant. Its numerical value is 0.5772... . It is not known whether 7 is rational or not.) = 10™ satisfies sy > 100? (0) Roughly how large must m be so that J we Solution. (a) We observe that log(N +1)—log N = f Fat, so that (sw4i— Iw log(N +1) — (sw ~ log N) = ghz — fy"? bat < 0. Thus the sequence is a decreasing sequence. On the other hand, it consists of positive numbers, since134 CHAPTER 8 SOME SPECIAL FUNCTIONS N 1 1 1 tog = [ qa “ Sy. " = i < e( function /(z) = (1 — 2)e%* has derivative (1 - 2z)e*, which is positive on this interval. Hence the smallest value this function has on the interval is its value at © =0, which is 1. We have now established the inequality for any integer N less than ps. tends to 00, so does the left, . Since the right-hand side of this inequality Exercise 8.11 Suppose f € R on | x — +00. Prove that Al for all A < 00, and f(z) = 1 as lim, * f(a) de =1 (t>0). 2 Jo Solution. We first observe that the improper integral converges absolutely for allt > 0, since 1) 59, [etitolars Bem JR t where M = sup [f(2)|, as R,.S = 00. 22R We also note that frets [err(eau and this last improper integral also converges for all t > 0. Hence we have fb [ere ae | = | [evr au [oug) le | “136 CHAPTER 8. SOME SPECIAL FUNCTIONS Since f(z) has a limit at infinity and () is Riemann- integrable on [0, 1], it follows that f(«) < K for some constant K and all x. Thus for any 7 > 0 we have tf ee) ée- | c+ Prertans [|r -1ldu < K+) +My TN 1], Hence, given ¢ > 0 we take n = gy. We then where My,e = sup |f (2) a choose X > 0 so large that |f(z) —1| < § if 7 > X, and we let =. It then follows that My, < § if0 sin?(n6) m5 nl (@) Let 50, and prove that + [eae (e) Put 6 = 7/2 in (c). What do you get? Solution. (a) Since f(z) is an even real-valued function, it makes sense to use the real form of the Fourier series, since symmetry shows that b, = O for all n. Then : ae Band form > 1 we have og == [ fla)cosnade = = [ cosne dz = 2Qsinné a9 77, (b) Since f(z) satisfies the Lipschitz condition of Theorem 8.14 at x = 0, it, follows that the series actually converges to (0) at that point, ie.,137 so that S sin(né) 7-6 am aot (6) Parseval’s theorem now implies that — “ite aifas=3()° + yo Sed) ant Now multiplying both sides by 3 gives the required result. (d) Let R be any fixed number, N any positive integer, and let éy = 5. As T6y = int R sinzy? N00 me have J SEAM) _, [" (SRE)? do, since the lot hand side of = 0 this equality is a Riemann sum for this integral. Note that ee Noy ~ (The inequality results from the fact that [°° ge < fe2, A dt = phy.) Given €, choose R > § such that | f° rsiney? S sina? fy e- LS) if S> R. Then choose No > 3 so large that ne R (ni) sin)? lo ae a (F) Cj ‘net whenever N > No. Then for N > No, 6y = § we have Ese-peye 0, and ap = 27. Compu- tation shows that b, = =2. Hence Parseval’s relation gives (2m)? wd, nal so that There is another way of deriving this result. Since = 1 Al x (Qn)? 4 _ me denoting this last sum by X, we find that Exercise 8.14 If f(¢) = |2|)? on (-7,7], prove that Led + Seosne = and deduce that — Cm a1 a Lae Lew (A secent article by E. L. Stark contains many references to series of the form S>n-*, where s is a positive integer. See Math. Mag., vol. 47, 1974, pp. 197-202.) Solution. Since f(z) is an even function, by = 0 for all n. The an’s are computed in a straightforward manner: a= 2 [fede139 ona? fo (n= 2) cosne de = card [ 2% cosnz de, b so that, eventually, we find a, = 4. This gives the stated Fourier oe, and since f(z) satisfies the Lipschitz condition of Theorem 8.14, the series converges to f(z) at every point. Taking © =0 gives the first of the two desired equalities: Parseval’s theorem yields Qt a a2 [inode which easily transforms to the desired relation. Exercise 8.15 With D,, as defined in (77), put Prove that. A (+1) ~ cos(N + Ve Kul) = FEIT cose and that (a) Kn 20, OF [ wwe (© Knle) < 74 gif0<8<|al f(a) uniformly on [-x,7}.140 CHAPTER. 8. SOME SPECIAL FUNCTIONS Hint: Use properties (a), (b), (c) to proceed as in Theorem 7.26. Solution. Using the formula 1 — cos@ = SRG42)E wwe deduce that in? $8, and the formula D,(x) = Nx a 1 (1 = cost) Ky (a) = maiy™ pesin(n + 5). 1 Now, however, sin asin = } cos(a — 8) — cos(a +), so that Se (cas(nz) ~ cos((n + 1)z)) = azo (1 = cos2)Kw(2) = (a) The nonnegativity of Ky(z) is an immediate consequence of either of the formulas just written. (0) It was established in the text that ef Dn(z)dz = 1, and so the same result for Kiv(2), which is an average of the Dn(z), must follow by routine computation. (@) This inequality is an immediate consequence of the facts that cos( + 1)z = —1 and that cosz is decreasing on [0,7]. The formula for oy(f;2) is an immediate consequence of the definition of ow(f:) and the corresponding formula for sa(f;2). Now let M = sup|f(z), the supremum being taken over all x. By (a) and (8) we have Few(e)— seal = [ge ite ~~ seonxew(nee s gf e-9-seiKnoa A 7 a | le 0 - Heinle) ae 1 1? +a ORT Teed S< sup |f(e-t) - f(a) + Q ise N+| | | 141 Mens) where Qs = xvirii=eorss . Given © > 0, we first choose 6 > 0 so smell that sup [f(e~#) ~ f(2)| < $ Iisé for all x, With this 6 fixed, we then have |ow(f;z) ~ f(z)| 22 and all « Exercise 8.16 Prove a pointwise version of Fejér's Theorem: Uff ER and f(a+), f(e-) exist for some x, then simowlia) = ZU let) + le Solution. We need only a slight modification of the argument just given, namely the formula ) ow( tsa) = $UUC@H) + $e =x [Wee 9- He Nkw Odes 5 [le 9~ Hen IKnlo ae Each of these two integrals can be broken up into an integral over a half neighborhood of 0 and an integral outside that neighborhood. The first of the integrals can be made small if the neighborhood is taken small enough (independently of NV). With that neighborhood fixed, the second integral in each case can be made small if N is large enough using the same inequalities just stated. Exercise 8.17 Assume f is bounded and monotonic on [-7,7) with Fourier coefficients cn, as given by (62). (a) Use Exercise 17 of Chap. 6 to prove that {nc,} is a bounded sequence. (0) Combine (a) with Exercise 16 and with Exercise 14(e) of Chap. 3, to conclude that (f(z+) + Fe-)] win lim sy(f;2) = for every = (0) Assume only that f € R on [-1,77] and that f is monotonic in some segment (2,8) ¢ [-",7]. Prove that the conclusion of (b) holds for every x € (a, 8). (This is an application of the localization theorem.) Solution. (a) by Exercise 17 of Chap. 6 we have sf. Slayer de = af 2 ay(z), from which it follows that Inca] < SUF) = f(A)142 CHAPTER 8. SOME SPECIAL FUNCTIONS (b) Since (2+) and f(x—) exist at every point, it follows from the previous exercise that ow(f;x) — 3[f(c+) + f(z—)]. Then Exercise 14(e) of Chap. 3 assures us that Sn(f;a) has the same limit. (0) Let g(x) = f(z) fora < 2 < B, g(x) = fla) for0 <2 ¢ — 144 tan’ 2, which is negative on (0, $). Hence all of the first six derivatives are negative on this interval, and therefore the function itself is negative. The same technique applies to the second function. All of its first five deriva tives vanish at x= 0 and the fifth is —[16sin 2c + 16 + 80 tana + 1362 tan? 2 + +200 tan? x + 240z tan‘ 2 + 120tan®« + 120z tan® 2), which is negative on (0, §). Hence this function is always negative on that interval. Exercise 8.19 Suppose f is a continuous function on RY, f(z + 2r) and @/7 is irrational. Prove that F(z), for every 2. Hint: Do it first for f(x) = ef.143 Solution. Following the hint, we observe that both sides of the desired equality equal 1 trivially when k = 0. In any other case the right-hand side is zero, and the left-hand side is ae 1 ell tho Nac Nie)? which tends to zero as N — oo. Since both sides are linear functions of f, it now follows that the relation holds for all trigonometric polynomials. Finally, since both sides are bounded by the supremum of f, given ¢, we can approximate f uniformly within e by a trigonometric polynomial. It then follows that all the means on the left, for NV sufficiently large, are within 2e of the integral on the right. Since ¢ is arbitrary, it follows that the limit on the left equals the integral on the right. Exercise 8.20 The following simple computation yields a good approximation to Stirling’s formula, For m= 1,2,3,..., define J(e) = (m+1=2)logm — 1m) log(m + 1) ifm 1 and that S(z) de = log(n!) - Ltogn > -4 +f (2) de, 1 2 ei Integrate log « over (1,n]. Conclude that 7 1 id 1) —(n+2) togn+ 3 < losin!) (n+ 5) togn nel for n = 2,3,4,.... (Note: log V2 ~ 0.918... .) Thus js nl OS Trea Se Solution. We first draw the graphs of f and g in the range 2 = 1 tox = 10. We note that f is merely the broken set of chords joining the points on the graph of logs at integer values of x, and g is made up of segments of the tangents at these points (g is not continuous), Because the downward side of the graph of log is convex, f(x) < logs < g(2) for all 2. The estimate for the integral of f is straightforward: The integral is the sum of the areas of one triangle and n —2 trapezoids with base 1 and parallel sides log and log(k-+ 1) (k= 2,...,n~1).44 CHAPTER 8 SOME SPECIAL FUNCTIONS We find it equal to }log1 + log2 +log3-+---+log(n~1) + }logn = log(n!) - 4logn, as asserted. "Meanwhile the integral of g(z) is also a sum of trapezoids and two triangles and equals } + log(n!) - }logn — 2. Hence we have 1 a 1 a dee 1 g(n!)—2 dL ptog(nt)—togn- 2 < b i log(n!) glesn < [ loge de < 3 +los(n!)—5 logn—5- < 5 +og(n!)—5 logn. Now straightforward computation reveals that [ logs dr = (nlogn —n) — (Logi — 1) = nlogn—n+1 h The desired inequalities are now deduced by taking exponentials of the three expressions. Exercise 8:21 Let [Dalat (n =1,2,3, Prove that there exists a constant C > 0 such that Ln >Clogn (n=1,2,3, or, more precisely, that the sequence 4 {Ln ~ Ge n} is bounded. Solution. We observe that a1 [PH sinfnt be a att har (ciisinn + BM a a We (-1)"sin(n + 3)t sin #t mS gos t dt The substitution u = (n+ 3)¢ changes the first and last terms into the sum +i" (-1)"sinw Lie Hi The first of these terms tends to — ml sinudu = = as n — 00. The second tends to 0 (for w € [n, (n+ 3)x] we have sin( x) > 3 sin 28, a8), which tends to Las n— 00).‘Thus we find that 1 S 1 =te.4+ > — boast e| tast where é, 0 as n= 00. If we take out the first two terms of the sum instead of just the first, we find similarly that C [sin(n + })¢] dt+ 0 2) ol (cay sin(n i aed (ay sini + Bt mat Jak sin jt : Again the substitution u = (n + 4)t changes the first and last terms into the sum rie (-1)%sinw aus f wae) du, ef 2 terms tends to = [ |sinu|du = = as n + co, and once lo = im again the second tends to zero, Thus we find that where tn 0 as n — 00. Once again, in each of the integrals under the sigma in the last two inequal- ities we make the substitution u = (n+ 3)t. When we do so, we have 2 2 =o a Sdn Si tm ty ae rsin( EP) 7 iat where €, + 0 and 7, — 0. It therefore follows that S2 1 1 Zz = for0 0.| 447 Solution. Following the hint, we use the ratio test to establish that the radius of convergence of the power series that defines f(z) is 1. This amounts merely to the statement that The differential equation then results from termwise operations on the series and the fact that a(a=1)--(@=n) , a(a=1)- (ant) (a-n+1) nl @-i nie ‘Then, given that f(0) = 1+ 0, it follows that for ¢ near 0 we have £@) _ _@ f(z) 142° so that log f(z) and log(1 +.2)* have the same derivative, and hence differ by a constant, which turns out to be zero, since both equal 1 at x = 0. It thus follows that f(z) = (1+) To prove the other relation, we merely observe that (-a(-e-1)---(-e-n+1) = (-1)"a(a+1):- (o+n-1)= (yee) Exercise 8.23 Let +y be a continuously differentiable closed curve in the com- plex plane with parameter interval [a,b], and assume that 7(t) # 0 for every t € [a, 0]. Define the indec of y to be a2 yo “2m J, Ind dt. Prove that Ind (7) is always an integer. Hint: There exists ¢y on (a, 0] with g' = 7//7, ya) =0. Hence yexp(—y) is constant. Since +(a) = 7(b), it follows that exp(y(a)) = exp(y(b)) = 1. Note that y(b) = 2niind (7). Compute Ind (7) when (t) = e™, a = 0, b = 2n. Explein why Ind (7) is often called the winding number of y around 0. Solution. Again, following the hint leaves very little to do. We define =) ga)= | Yat, v= [58 so that we automatically have y'(t) = “2. The fact that yexp(—y) is con- stant is now a consequence of the chain rule. It then follows immediately that exp(y(b)) = 1, so that (b) = 2rin for some integer n. Routine computation shows that Ind (7) = n if y(t) = ei, 0 < t < 2n. Since this curve winds counterclockwise about 0 a total of n times, the name winding number is appropriate.148 CHAPTER 8. SOME SPECIAL FUNCTIONS Exercise 8.24 Let + be as in Exercise 23, and assume in addition that the range of 7 does not intersect the negative real axis. Prove that Ind() = 0. Hint: For 0 < ¢ < 00, Ind (7+) is a continuous integer-valued function of c. Also, Ind (y +c) +0 as ¢ + 00. Solution. Following the hint, we observe that 1 ve HO = 53 f WO +e is a continuous function of c on [0, 00), since Ila) - @) 0 and 0 0 such that |y(t)| > 6 for all t € [0,2n]. If P and Py are trigonometric polynomials such that |P;(t) —7(t)| < 6/4 for all t € (0, 27], (their existence is assured by Theorem 8.15), prove that Ind (P;) = Ind (P) by applying Exercise 25, Define this common value to be Ind (7) Prove that the statements of Exercises 24 and 25 hold without any differen- tiability assumptions. Solution. Since |P,(t) — Pa(t)| < $< |Pi(t)|, (because [P,(t)| > [F(#)| — [F(t) — P,(t)| > %), the equality of the indices follows from Exercise 25, as stated. Exercise 24 remains valid, since if (t) does not intersect the negative real axis, there is a positive number 6 > 0 such that |y(t) — 2| > 6 for all z < 0. Then if |P)(¢) — 7(¢)| < 4 for all t € [0,2n], it follows that P;(t) also does not intersect the negative real axis, hence has winding number 0. Exercise 25 remains valid, since if |74(t) —9(t)| < /ni(t)| for all ¢, we can let 6 = ming /7x(¢)|- /n(¢) — 72(6)]. Then if |P:(t) — 74(t)| < 6/4 for alll, it follows that |Pi(t) — Pa(t)| < Ina(t) ~ va(¢)| + (6/2) < In @I— (6/4) < |Pi(®)], and so Ind (P;) = Ind (Pa), by Exercise 25. Exercise 8.27 Let f be a continuous complex function defined in the complex plane. Suppose there is a positive integer n and a complex number ¢ # 0 such that lim 2°"4(z) = 0. Prove that f(z) = 0 for at least one complex number z. Note that this is a generalization of Theorem 8.8 Hint: Assume f(z) #0 for all z, define alt) = F(re) for 0 0. $(0) for all t, we have 94(t) = 0 for all ¢, and hence Solution. (a) Since yo(t) by definition Ind (70) (8) Choose R so large that |2-"/(z) — cl < / whenever [z| > R. Then for all r we have Ind (7) = Ind (7-1) + Ind (7,2), where 741(t) = reM* and150 CHAPTER 8. SOME SPECIAL FUNCTIONS Figure 8.1: The Brouwer fixed-point theorem alt) ‘e-int f(reft), By Exercise 25 we have Ind (7-2) = 0 for r > R, and by direct computation we have Ind (71) =m for all r. (c) Fix ro > 0, and let ¢ = ymin, |f(roe| Then choose 6 € (0,70) such that [f(roe'*) — f(ret*)| se for n = 1,2,8,... . Use Theorem 8.20 and Exercise 30 to show the more precise result, 1 lim vif (l—2?)"de = yz. aaa fy Solution: Let u in the integral, so that dz = 4u7? du. We then have adda Vin + DEE) Tm+l) va f.a-2rae= va f'0- and taking ¢ TQ) = in Exercise 30, we find that this last expression tends to