Solutions Manual A First Course in Chaotic Dynamical Systems: Theory and Experiment by Robert L. Devaney Thomas Scavo scavo@cie.uoregon.edu January 27, 1993Contents List of Figures Preface 3 Orbits 4 Graphical Analysis 5. Fixed and Periodic Points 6 Bifurcations 7 The Quadratic Family 9 Symbolic Dynamics 10 Chaos 11 Sarkovskii's Theorem 12 The Role of the Critical Orbit 13 Newton’s Method 14 Fractals 15 Complex Functions 16 The Julia Set B 19 37 59 at 93 109 125 133 163 110 195, CONTENTS ii 17 The Mandelbrot Set 209 A Mathematical Notation 229 B Map Index 2arList of Figures Bu 32 33 34 35 36 37 38 39 4a 42 43 44 45 46 4a 48 49 4.10 4d 42 413 414 445 4.16 ‘The absolute value function A(s) = ‘The graph of F(z) = 27+ 1 is a parabola with no fixed points, The graph of the cubic equation F(z) = 2° ~ 32, ‘The graph of F(z) = 2sinz. Graph of F(z) = [2-2]... - ‘The doubling map D(z) = 2 mod 1. ‘The second and third iterates of the seting map. ‘The tent map T(z) =1 ‘The second and third fe— 1). . erates of the tent map. ‘Two cases of linear maps with repelling fixed points. . ‘The graph of F(2) = zh, a typical rociprocal function A pair of quadratic functions. The graph of the cubic F(x) =~. ‘The graph ofthe sine function. . ‘The dynamics of two more quadratie functions. The dynamics ofthe ent map Us) Phase portraits ‘The straight line (x) ‘The graph of F(z) ‘The graph of F(z) =z"... ‘The exponential function B(z) = 6. | Orbits of the slide-and-fold dynamical system F(z Linear maps of the form F(z) = fixed points. inear maps with repelling fixed points. Straight lines parallel to the diagonal. — [22-1]. |. 30 LIST OF FIGURES 47 ba be 53 54 55 56 57 58 59 61 62 63 64 6.6 67 Ta 72 10.1 10.2 10.3 10.4 ua 2 13 4 5 Straight lines perpendicular to the diagonal. . ‘The doubling map D(z) = 2x mod 1 : Examples of neutral fixed points which are weakly attract- ing on the left and woakly repelling on the right. ‘The graph of F(e) = 1/2, a totally periodic function. ‘The unit circle in the first quadrant Examples of neutral fixed points which are also inflection points. Such pins ar either wey rpeng or wally attracting Neutral fixed points with a detivative of —1. Two cases of a neutral fixed point. ‘Two more cases of a neutral fixed point. . ‘The Four Canonical Forms of Neutral ied Points, Bifureation behavior in the quadratic family F\(2) = 2+ 24d, Bifureation behavior in the cubic map Fy(z) = At-+ 2°, ‘Typical mombers of the sine family Sp(2) = pain e. ‘Two examples from the exponential family By(z) = M(e*—1). Every member of the family H_(2) = 2-+ cr? has a neutral fixed point. Representatives of the family F.(z)=2 + ex? +24 Bifurcation diagram for the logistic family. ‘The graph of T(2) = 3/2~ [Sx — 3/2]. ‘The second iterate of T(2) = 3/2 — |S ~ 3/2 A binary search algorithm for w. A binary tree of the elements in So. The quadratic map Q-2(z) = 2? — 2 and its piecewise linear approximation V(x) = 2|2|— ‘The second and third iterates of V(z' 2a ~ 3 A discontinuous piecewise linear map having only period 3 points. Bieda tee Two graphs with period 4... This funetion has a 7-eyele but no 5-cycle. ‘Two graphs with no odd periods. . A contimuous function with a 3-cycle, 100 101 110 112 14 115 uy16 wT 118 BA 132 133 134 135 13.6 13.7 138 139 13.10 13.11 13.12 1 13.13 13.14 13.15 13.16 13.17 188 13.19 13.20 13.21 13.22 Ma 42 143 Ma LIST OF FIGURES A subshift of finite type on three symbols A subshift of finite type with a Seyele but no fixed or period 2 points. . = eeuvageeg More subshifts of finite type. ‘The Newton LF. for F(2) = 4 ~ 22 is constant. : ‘The Newton LP. for F(z) = 2% — 22 is a typical ample The Newton LP. for F(z) = 2° is linea. ‘The Newton LF. for F(z)=24+27.... ‘Two members of the family F(a) = jee.) The Newlon LF. for F(s) = /VI-+ 24 a simple cubic polynomial The Newton LP. for F(x) =e". 2... ‘The Newton LP. for R(x) = / i linear. ‘The Newton LF. for F(x) = #/(z — 1) is the parabola preet : ‘Two more members ofthe family F(2) ~ e/(e ~ 1)" A distant view of F(z) = 2/(z - 1). ie Newton iteration neti for (=) = 3° fs linear, ‘Typical members of the family F,(2) = 2® + p and their associated Newton iteration functions. Dynamics of NF. anne Chaotic dynamics of NF; for > 0. a tH HE ‘The Newton iteration funetion for Go(2) = 2x ~ 1). ‘Typical members of the family Gy(z) = 2°(2—1) +m and their associated Newton iteration functions. . . . . Dynamics of NGo(2) = (#(2e — 1))/(32 ~ 2) ; A quartic polynomial and its associated Newton LP. The Newton iteration function for F(s) = =(244+1) ha 8 globally attracting fixed point at the origin. . . A pair of Newton LPs for F(x) = (2? ~ 1)(2? + 4). ‘The Newton LF. of G(z) = exp(-1/2%) has a way at tracting fixed point at the origin. ‘Two iterations of Equation 14.1. . 2... ‘The first stage in the construction of the Cantor middle. fifths set serseeseete ‘Three iterations of Equation 14. A Kocblike transformation, 19 120 121 134 IM 135 136 137 139 140 40 2 143 4 ar ur Mg M9 151 151 152 154 156 158 160 164 165 167 172 LIST OF FIGURES 5 146 147 148 149 15.1 15.2 172 173 Wd 175 176 ‘Three iterations of a Koeh-like curve. ‘The fourth iterate of the transformation in Fig. 14.4, ‘A coloring of Pascal's triangle ‘Another coloring of Pascal’ triangle Yet another colosing of Pascal’ triangle. The orbit of 1 under iteration of L(z) = és/2. ‘The orbit of 1 under iteration of complex lineat maps a repelling fixed point at the origin. . . . ‘The orbit of 1 under iteration of complex linear maps with a neutral fixed point atthe origin... . ‘Am affine map F is conjugate toa linear map L via another affine map H ‘Two illustrations of the Boundary Mapping Principe. th ‘The boundary of the set of e-values for which Q. has an attracting fixed point is a cardioid. Quadratic maps in a uighborood ofthe p Lifureation point © = —3/4 A bifurcation diagram for Qe. Representatives of Gy in a neighborhood of A A bifurcation diagram for Gy. The boundary of the fixed point region for FF, ‘doubling ‘Complex conjugates are symmetric with respect to the real Fixed point regions for various values of d. ‘The graphs of f and g intersect at z= em! |. ‘The evolution of the fixed point region as d varies between and3.. 2... More geometry of complex numbers. ‘The point z = is a Misiurewies point 222 173 174 175 176 17 183 184 185 187 194 200 201 203 204 205 207 au 218. 219 221Preface ‘My hope is that this manual will benefit teachers and students alike, and 50 most of the solutions have been written in great detail, with many inter- ‘mediate steps, thereby avoiding a common complaint anioug mathematics students. An instructor might use them to supplement his or her lecture notes, and students will certainly want to compate them with their own handwritten solutions The manual is reasonably complete. However, all of the chapter 8 exer- cises (iuost of which ate experiments) have been neglected, as well as the ‘material in chapter 18. Of the remaining exercises, approximately 330 have been solved. A few supplemental exercises have been added at the end of chapters 3 and 4, and to make the analysis of neutral points more straight- orward, some results notin the text have been added to chapter 5. There are two appendices: list of mathematical notation and a Map Index (that is, an index to all of the mappings in the text itself). The solutions manual was typeset using OATpX, Andrew Trevorrow's ex: cellent freeware TX implementation for the Macintosh. Most of the figures were done with PSMathGraphsiI, and I’m grateful to the program’s author, Jol Jacob, for adding an Orbit command at my request, without which ‘many of the figures, especially those in chapter 4, would have been hope- lessly difficult. I've also benefited from the productivity and generosity of many individuals, especially Larry Siebenmann, Pablo Straub, Pete Kele- her, Steve Cochran, Rick Zaccone, Cameron Smith, Piet van Oostrum, and Paul Taylor, whose macro packages and applications I've come to depend Most of what I know about dynamics was learned at the Regional Insti- tute in Dynamical Systems at Boston University over the last three sum- ters. ' like to thank the conference organizers and the National Science Foundation for making it all possible. I'm particularly indebted to my PREFACE vil friends John ‘Tho, Larry Riddle, and Rod Shaughnessy for valuable dis- cussions while at RIDS. But most of all, ’d like to thank Bob Devaney, who (although he may not realize it) continues to be a tremendous source of inspiration and support ‘Tom Seavo Eugene, Oregon January 1993Chapter 3 Orbits Exercises 1. Let F(2) = 2%. Compute the first five points on the orbit of 1/2. PQ)=F@)=4 PQ)= FQ) = 1 Pd) PG) = Fis) = aise FAQ) = F(ba) = aoa (Note how these fractions visually trace out a parabola!) It appents that rn# in general, and this can be proved by induction (see Exercise 5) 2. Let F(x) = 2% +1. Compute the first five points on the orbit of 0. 10) = F(O)=1 FX0) = FQ) =2 P0) = F(2) =5 FM(0) = F(6) = 26 (0) = F(26) = 677 Exercise 5 2 (See Figure 3.2 for the graph of F.) 3. Let F(2) =~ 2. Compute F%(2) and F°(2). ‘The second iterate of F is F(z) = F(F(2)) = (2? 2) -2 while the third iterate is given by F(x) = F(F%(2) wt ds? 42 (24 4? 42) A 42? 42)?—2 — 82° + 2024 — 1627 42, Fis a lengthy 16th-degree polynomial with nine terms. It appears that P* isa 2"th degree polynomial with (2-1 +1) terms, What patterns do you see in the iterates of F? 5. Let F(2) = 2. Compute F%(x), F%(2), and Px). What is the formula for F"(2)? We have that FH) = FUF(@)) = FG) and P(x) = F(P%(2)) = P(e") We also have F(z) = F(F(a)) = FG) In general, it appears that F(a) = 2 (3.1) which we will now show by induetion (sce Exercise 1 for a special case). We've already verified that (3.1) holds for n = 1, 2, 3, and 4, and so the base case of the inductive argument has been established. Now, suppose that the equation is true for n := &. Then PEA (a) = F(PM(2)) by the inductive hypothesis3 CHAPTER 8. ORBITS Figure 8.1: ‘The absolute value function A(z) = [2]. which proves that (3.1) holds for all n. Note how this argument mirrors the above computations of F%(x), F%(x), aud F4(x), by Uhe way. 6. Let A(z) = [2]. Compute A%(x) and A%(2). By definition, a. Fadi a@=tl={*, P22). and 90 A(x) 2 0 for all = (sce Figure 3.1). We also have that A*(2) = A(A(2)) = |A(@)| = AC) since A(z) > 0. Similaely, AX(2) = A(A?)) = A(A(2)) = [AQ@)|= AC), and in fact, AM2)= AG) for n > 1, (Can you prove this by induction?) What does this mean? Tt implies that A hae no petiodic points of prime petiod n > 1. 7 Bind all real fixed points (if any) for each of the following functions: Ta) F(e) = 8242 Be$2= Exercise 7 4 z?-2=2 = 2?—2—2=0. Applying the quadratic formula to this, second degree equation, we got pit VI=MC _ 14-v5 2 z (-1,2} Te) F(2)=27 +1 (soe Figure 3.2) e4loe so etaet we get _ 1k YI= a) _ 4 vii 2 2 In this case, the fixed points are complex and the reader is referred to Chapter 15 for details concerning complex functions. Td) F(z) = 29-32 (see Figure 3.3) P3223 B-42=0 = a(z?—4)=0 2=00re= 42. ‘Consequently, fix F = {0,42} 1) P(e) Since and therefore, fix P= 0. Again applying the quadratic formula, le] (see Pigure 3.1) = 2 for nonnegative 2, we have that fix F = {2|2>0).5 CHAPTER 3. ORBITS Figure 8.3: The graph of the cubic equation F(z) = x* 7) F(a) =2* > aba 0 = 2(z*— 1) =0 which has the fairly obvious and # =41. Rut thera are alan two ather (complex) solutions. Can you find them? 7g) F(z) = 2° a = 28-2 =0 = 2(z-1) =0 which has solutions 2 = 0 and 1, and four others which are not so easy to find (see Chapter 15), Note: ‘The solutions to 2” ~1 = 0 are called the nth roots of unity. 7h) F(z) = 2sinz Observe that 2 230 sing = 2(1—sinz), Thus, ee Dori —snz <0. Now, shz= ite 1/2 or any Be-mlipl or3/2. Tham fan Font tant fab ={.., Ete Btn f Zate En FU) Sr x Sr or =f EES duo = {(4k+ 1)x/2|k € Z}U{O}. Note: F is an even funetion, that is, F(2) = F(—2) for all z. See Figure 3.4. 8. What are the eventually fixed points for A(s) = [el? Exercise 10 Figure 3.4: The graph of F(x) = xsin x {As shiown above in Exercise Te, fix A = {2 | 2 > 0). But fixed points ace eventually fixed with preperiod 0. The negative real numbers are also ‘eventually fixed since each becomes positive after just-one iteration. ‘Thus, all real numbers are eventually fixed under iteration of A. We write TkA=R to denote this fact. 9. Let P(x) = 1-27. Show that 0 lies on a 2-cyele for this function, ‘This is most certainly true since (0) = 1 and F(1) = 0. 10. Consider the function F(a) 2-2if2>2 n= (SEE See Figure 3.5 for the graph of F.! 10a) What are the fixed points for #? Iz > 2, then jo—2)=2 = 2~2= 2, which has no solution. On the other hand, if < 2, we have that |r~2|=2 =» 2— 2=2e = 2=1. Therefore, fix F = {1} walled the alias i-fold dynamical system on pp. 141-142 oft Barns ley; Michael. Fictals Bverywhore, Banton: Academic Pres, 1988,7 CHAPTER 8. ORBITS Figure 3.5: Graph of F(2) = |2 ~ 2h 10b) If z is an odd integer, what can you say about the orbit of #? Observe that if 2 is an odd integer, then so is x 2. Now, suppose in an odd positive integer greater than 2. ‘Then 2 —2 is an odd positive integer smaller than 2, and repeated subtraction eventually Produces a value of 1 which is fixed by #. In other words, all odd Positive integers are eventually fixed. But what about odd negative 2? Well, since #*(z) > 0 for all z, all we need to do is apply this very same argument to F(2). 10) What happens to the orbit if x is even? Suppose 2 is an even positive integer greater than or equal to 2. Then 2 is an even positive integer smaller than z. In this case, repeated subtractions eventually vanish, but (0) = |0~2| = 2. And since F(2) = 0, we see that the orbit is eventually canght in a eycle of period 2. Similar arguments hold for even negative x ‘The following four exercises deal with the doubling function D. 11. For each of the following seeds, discuss the behavior of the resulting orbit under iteration of D. La) 20 = 0.3 Since D(0.3) = 0.6, D(0.6) ‘D(0.8) = 0.6, the orbit of 0. 0.2, D(0.2) = 0.4, D(0.4) = 08, and eventually periodic with preperiod 1 Exercise 12 8 and period 4. We write 0.3 © per} D. Lb) 20 = 0.7 Since D(0.7) = 0.4, and since 0.4 € per D from Exercise Ma, it follows that 0.7 € pee} D. Le) #0 = 1/8 1/8 ~ 2/8 + 4/8 r+ L mod 1 = 0. But 0 is fixed by D. ‘Therefore, 1/8 € per} D C fixD. 1d) x = 1/16 1/16++ 2/16 ++ 4/16 + 8/16 ++ 0. Therefore, 1/16 € per} D. Me) 20 = 1/7 Since D(1/7) = 2/7, D(2/7) = 4/7, and D(A/7) = 1/7, we have that 1/7 € per D. 1if) 20 = 1/4 Note that D(1/14) = 2/14 = 1/T. But it was shown in Exercise He that 1/7 € perg D. Therefore, 1/14 € pet} D. Lg) 20 = 1/11 Since 1/11 e+ 2/1 + A/T Hy 8/LL 4 8/1 ms LO/LL Ofte ‘U/L + 8/1L++ 6/11 1/11, we see that 1/11 & peryo D. Uh) 29 = 3/22 Since D(3/22) = 6/22 = 3/11 € pery D by Exercise 1g, we have that 3/22 € perty D. 12. Give an explicit formula for D(z) and D%(z). Can you write down a general formula for D"(z)? Recall that the doubling function D is given by the equations D(z) = 22 mod 1 afm ifo D%e)=D(De)) 12<2<3/4 > O D2) SASe Y2 D%2) ‘Thus, we have shown that 4r if 0 0 1/4< D(z) <1/and 1/2< DYz) <1 + DY@) = DID) Diaz) 242) —1 =8r-1 Can you predict the outcome? ‘The reader is encouraged to complete the remaining six cases. When the dust clears, you should get Be if0 2=0 4e-1=2 > 2=1/3 4e-Qe2 > 2=2/8 4o-3=2 > 221 (mum. ..,Phis seems to be saying that x = 1 is a fixed point of D?, Is this « true?) Similarly, we see that fix D® = per D = {0,1,3,3,4,,$) sinee 8r=2 > 2=0 8r-ls2 > 2=1/7 8e-2=5 + r= 2/7 8e-3a2 > 2=3/7 so ra4f7 8r-ba2 > 2=5/7 fr—6ae r= 6/7 8e-T=2 > r=1 (Once again we arrive at the solution x = 1. Maybe we should define D:[0,1] —+ [0,1] with D(1) = 1 for consistency?) From Equation 3.3, we see that fix D" = per, D={k/(2"—1)|0 W2STR)<1 > TH W2 2KT(e) T%2)=T(T(2)) T(2—22) 2-2(2— 22) 42-2 T(T(2)) =T(2— 22) 22-22) 4-40 BASe O THe) ‘Thus, we have shown that 42 if 0 0 implies that *(z) > 0 for all n. Likewise, x <0 implies that F(z) <0 for all n. Thus orbits are destined to remain in their quadrant of origin for all time (see Figure 4.2). Secondly, not only is F an odd function, but it’s also its own inverse. Finally, an easy calculation shows that fix F = {43/3}, and that F(F(2)) = # for all 2, that is, pera F = R. The funetion F is an example of what we call a totally periodie function ‘These results may be generalized somewhat. Let F(2) = k/s for any £0. It can be shown that fix = {VE} and that F%(z) = 2 for all x "Am edd fonction le one for which F(~2) {imagine folding the graph along the parte ofthe curve ie one on top the oth =F(z) forall s. Geometrically speaking, ‘and then along the yraxi, 60 that the two Exercise 1 20 (2) The graph of P(e) = 22. Figure 4.1: Two cases of linear maps with repelling fixed points. Figure 4.2: ‘The graph of F(z) = gb, a typieal reciprocal function.a CHAPTER 4. GRAPHICAL ANALYSIS ye (8) The graph of Fle) = =~ 27 Figure 4.3: A pair of quadratic functions, (4) The basic quadratic F(2) = 24 le) F(z) = 2241 (see Figure 4.10) We have that fix F = {1/3} since dep lae > 1a82 > 2H 1/3, It’s also truc that |£(2)| + 00 for all x except the fixed point, Note the qualitative difference between this linear dynamical system and its cousin in Exercise 1a; that is, as the orbit moves away from the fixed point at 2 = 1/3, it begins to alternate sign and continues to do so for all time. 1d) F(e) = 2? ‘This map was briefly discussed in Section 4.3 of the text where it was pointed out that fix P = {0,1} and that —1 is eventually fixed, In = TRF = (0,41). Now, suppose 2 is nonnegative. ‘Then the following statements are true i. Ie > 1, then P®(z) + 00.08 n + 00. fi 10 <2 <1, then P%(2) +0 as noo Exercise 1 2 iil Ix = 1, then #*(1) = 1 for all n (that i, 2 = 1 isa fixed point for Fas we've already seen). But what about negative 2? Once again there are a number of eases to consider: iv. Itz < 1, then F(z) > 1. Hence, by (i) above, F"(2) —+ 09 as n—002 v. Now, if -1 < x <0, then -z > 2” > 0. (This is obtained by multiplying through by z, which is negative.) But —2 <1, and 80 we have that 0< 2? <1, Hence, F"(e) +0 as n — oo by (ii) above. vi, When z = —1, we see that F(z) = 1. But 1€ fixF and soz 1 is eventually fixed. ‘Summarizing these results, we write 0 iffel1 We remark that the behavior of /°"(2) for negative 2 is analogous to that for positive z since one iteration of F maps its argument directly into the positive real numbers (i.e. F(z) = 2? > 0 for all =). See Fig- lure 43a. The importance of this exercise can not be overemphasized. Je) F(a) = —28 Here's another odd function (see Figure 4.4), this time with fix = {0}. We also see that F2(2) = 2°, and we have that -t1 € per". ‘There are other period 2 pointa—the remaining 8th roots of unity—but these are complex. Anyway, it turns out that the fixed point is attracting while the 2-cycle is repelling (see To be predic, is the orbit of F(z) that tends to co. But if F*(F(2)) -+ 00, then 0 doce PM(2) since is the tal of the sequence tha determines convergence,23 CHAPTER 4. GRAPHICAL ANALYSIS \ Figure 4.4: ‘The graph of the cubic F(z) = Chapter 5), and so 0 it\e}<1 mvreain{ | thlSt 42) me co if |e|>1 Compare (4.2) with (4.1): the absolute value signs are required in the former since points along the orbits alternate in sign, 18) F(x) ‘As seen in Figure 4.3b, this function has a single fixed point at the origin. If 0 1, it follows that (2) < 0, and consequently, F°(F(z)) — —c0 as n+ oo. Finally, note that 2 = 1 is eventually fixed after one iteration. Experiments indicate that for 0 < 2 < 1, the orbit of x converges to 0 rather slowly. Similarly, for negative z very close to 0, F*(2) slowly ‘moves away from the origin. The explanation of this behavior is that the fixed point is neutral, an important characteristic discussed in more detail in Chapter 5. 1g) S(@) = sine Looking at Figure 4.5 it appears that all orbits tend to 0, whi Exercise 3 a Figure 4.5: The graph of the sine funetion. true, but experiments suggest that the convergence is very slow. (See Section 3.2 of the text for an illustration of this fact.) 2. Use graphical analysis to find {2 | *(2) — too} for each of the following funetions 2s) F(z)=22(1—2) (see Figure 4.6a) For 0<2< 1, P®(z) — 1/2. But when 2 < 0, F(z) + —00. Now, if > 1, it follows that F(z) <0, and so F*(F(2)) — —o0 as well. ‘Thus, {2 | PM(z) +00) = (2 |2 <0 ore > 1). Note that 2 is eventually fixed after one iteration. 2b) F(z) =2 +1 (see Figure 4.6b) In this case, we have that F"(2) + co for all af% ifesi2 29) n= { 2-22 ife>1/2 ‘The map depicted in Figure 4.7 is called the tent map. When x <0, T"(z) -» ~00. When z > 1, T(z) < 0, and so T"(T(2)) + 00 as ni 00 a8 well. But when 0 << 1,0< T(z) <1 for all n. ‘Thus, {2| PM2) -+ boo} = {| 2 <0 orz > 1). We remark that z=1 is eventually fixed after one iteration since T(1) =0, and 0 is fixed. 3. Sketch the phase portraits for each of the functions in Exercise 1.5 CHAPTER 4. GRAPHICAL ANALYSIS: (2) The graph of F(a) = 29(1~=)- Figure 4.6: The dynamics of two more quadratic functions. (0) The graph of Fs) = 22 41 Figure 4.7: The dynamies of the tent map T(2) = 1 — [2x — 1] Exercise 4 26 See Figure 4.8. 4. Perform a complete orbit analysis for each of the following functions ‘Thus, fix = {4}, and it so happens that F"(2) — —4 as n — oo for all x. See Figure 4 4b) A(z) = |e ‘This exercise was essentially solved at the end of Chapter 3 (see Ex: excises 6, Te, and 8) where it was shown that fixA={2|2>0), and that all other points are eventually fixed after just one iteration 4c) F(x) = —2? From the graph of F in Figure 4.10, we see that fix F = {—1,0). Since PQ) = =1, we also see that 2 = 1 is eventually fixed. Por -1 <2 <1, F(t) + 0 as n = 00, and for 2 > 1, F™(2) + ~00 a8 n+ 00. That is, O is attracting and —1 is repelling 4d) F(x) = —28 First, let's compute the fixed points of F: rte =a(lt24), =0 >0 ‘Therefore, 0 € fix F and the other four fixed pointa are complex. Now, Po A2 CHAPTER 4. GRAPHICAL ANALYSIS Exercise 4 (8) Fe) = 1/(2) has fixed points at 49/8, and all nonzero real numbers are of pet aS rt (©) F(a) = ~2e +1 also nas a globally (4) (2) = 29 has uwo types of orb repelling fixed point, but orbits eacillate those that approach 0 from the rg In thie case those that escape to oo. (©) Plz) = ~2" has a reeling 2eyee, (0) F(a) = =~ x4 has a neutral fixed (Compare with Figue-4.7 inthe text.) point which i weakly repalling on the Jet and weakly attracting onthe right. end oes (x) S(z) = sine has a globally, but weakly attracting fixed point at the origin Figure 4.8: Phase portraits. Figure 4.9: The straight line F(2) = 2/2 Figure 4.10: The graph of F(z) = 2829 CHAPTER 4. GRAPHICAL ANALYSIS Figure 4.11: The graph of F(z) and so 0 is superattracting since /"(0) = 0. Note also that 1 and —1 constitute & repelling 2-cycle for #. ‘This suggests that [F"™(2)] —+ 0 asm — co whenever || < 1. All other athita escape ta no, See Figure 4.11 40) Fle) = 1/2 Reciprocal functions of the form F(x) = k/x were solved in general in Exercise 1b where it was shown that all real numbers are of period 2 4) B(x) As shown in Figure 4.12 all orbits quickly escape to infinity under iteration of this exponential map. 5. Let F(2) = |z—2]. Use graphical analysis to display a variety of orbits of F. Use red (o display eycles of period 2, blue for eventually fixed orbits, and green for orbits which are eventually periodic. Recall from Exercise 10 at the end of Chapter 3 that fix = {1} and that all odd integers are eventually fixed. In fact, these are the only eventually fixed points for F. It was also shown earlier that {0,2} C perg F and that all even integers are ‘eventually periodie with prime period 2. In fact, per, F=(2|0<2<2}, Exercise 5 30 Figure 4.12: The exponential function E(x) = e* Figure 4.13: Orbits of the slide-and-fold dynamical system F(2)31 CHAPTER 4, GRAPHICAL ANALYSIS and moreover, every real number is eventually periodie with period 2. ‘To sce this, first suppose that x > 2 ‘Then there exists an m such that 0.< P™e) <2 (see the detailed argument in Exercise 3.10), and therefore, F™(c) € perp F. Consequently, x € perf’ F. In the event that x < 0, we have that, F(2) > 2, and so F(z) is eventually periodic via a similar argument. See Figure 4.13 for some typical orbits 6. Consider F(2) the fact that these pointe are also solutions of F2(2) of prime period 2 for F. We begin by computing — 1.1, First find the fixed points of F. Then use to find the cycle wad Ses which implies the fixed points of F are Lt VER 2 Just to be sure, we had better check our work: rei) (92) 1e2V544+54 4 4 6A 25444 a 24254 4 1s Voa ar] Next, let’s compute the second iterate of F°, Fe) 1h Set 2207 4191-14 = 2.207 40.11, v Exercise 7 32 and its fixed points (which are also the period 2 points of F): xt-292 40.112 > 2t-2227-24+O11=0 > (= 2-Lafet+2-0.1)=0 Sp rta2-L1=0 or a?+2—01=0 5 pe ltvSd |, -1tvia 2 2 But, you ask, how could anyone know how to factor a messy fourth degree polynomial such as that? Well, we know that any fixed point is also a period 2 point, right? ‘This means that the solutions to 27 — 2 — 1.1 ‘must also be solutions to x*—2.227—2-+0.11 = 0, and suggests we compute 2.282401 ap soy #21 by polynomial long division, say. (Thought you'd never use your college algebra, ch?) Study tis technique warefully—it will prove invaluable in the sequel. T. AlLof the follo of the form F(2) ing exercises deal with the dynamics of linear functions ie-tb where a and 6 are constants. 7a) Find the fixed points of F(x) = ar-+ 6. We have that artb=r provided a # 1. In other words, fix F = {b/(1—a)}. 7b) For which values of a and b does F have no fixed points? F has no fixed point when its graph is distinct from and parallel to the diagonal line y = 2, that is, when a= 1 and 6 0. Actually, a fixed polnt ls peviod n pofnt for any ns33 CHAPTER 4. GRAPHICAL ANALYSIS Bxorcise 7 4 @ocect Figure 4.14; Linear maps of the form F(z) = ax +6 with attracting fixed points. W)-1 1. Using graphical analysis, what can you say about the behavior of all other orbits of F in this case? We call such fixed points repelling. Can you explain why? Again, there are two cases to consider (see Figure 4.15): either @ > 1 ora < =1. Incither case, "(z)| — oo a8 n — 00, Note that the orbit, oscillates about the fixed point in the later case, growing arbitraily large in absolute value. Th) Perform a complete orbit analysis for F(2) = 2+ in case b> 0, 6=0, and <0, ib=o Here F is the identity map and fix F = R. i b>035 CHAPTER 4. GRAPHICAL ANALYSIS My Figure 4.16: Straight lines parallel to the diagonal In this ease, pery F = 0 for n> 1, and P*(2) + 00 as n = co for all 2. iib 1, but in this ease, and for all 2, F"(2) + ~c0 as nm — 00. See the sketches in Figure 4.16 for typical examples. “i Perform a complete orbit analysis for F(z) = —2+6: Regardless of 4, we find that fixF = {0/2} and per, F = R. See Figure 4.17 Here ate some additional exercises involving linear functions: Tj) Compute F? and discuss its dynamics. What is fix F? (and henee pet, F)? Under what conditions is this fixed point attracting? Re- pelling? Under what conditions does F? admit an oscillating orbit? Tk) Compute F° and F*, From these deduce # formula for F*. Prove by induction that your formula is correct. What is its limiting value as noo? Exercise 7 Figure 4.17: Straight lines perpendicular to the diagonal. 36Chapter 5 Fixed and Periodic Points Exercises 1. For each of the following functions, find all fixed points and classify them aa attracting, repelling, ot uvuural, 1a) F(x) = 2? — 2/2 8—2/2=2 > 2?—32/2=0 Therefore, fix F = {0,3/2}. P(e) = 21/2 F'(Q) = -1/2 = 0 is attracting and oscillating, PY(3/2) = 5/2 > 1 > 8/2is repelling, le) F(z) = 32(1—2) 32(1—2)=2 > 22—327=0 Exercise 1 38 + 2(2-32)=0 250 or r= 2/3 ‘Therefore, fix F = {0,2/3}- P(e) = 3-62. (0) =3.>1 = 0 is repelling, F"(2/3) = 3 - 6(2/3) =—1 = 2/3 is neutral Jd) Fle) = (2—2)/10 (Q-2)/l=2 > 2- => 21 ‘Therefore, fix F = {2/11} Piz) = -1/10. FQ/11) = -1/10 = 2/11 is attracting and oscillating, 42942 2 — 2, then Q*() = 24 — dz? 42 by Exercise 3.3, and Now a fixed point for @ is also a fixed point for F since Q"(@) = Q(Q(e)) = Q(x) = 2. But the fixed points of Q satisly the equation 2?— z—2= 0. This suggests we factor and solve the resulting 4th-degree polynomial equation as follows: atadt42=2 = tds? 42-0 3 (2-22? 42-1) =0 =0 or ba > (2-2e41)=0 oF a2 tev OED -ttV5 2 SP=2 or r=-1 or a= If we let then the reader may check that fix F = {—#,-1,2,1/6),39 CHAPTER 5. FIXED AND PERIODIC POINTS ‘The number ¢ is called the golden ratio! and arises naturally in connection with the ubiquitous Fibonacci sequence.? Now, since F@ iz Q(z), we have that F"(—1) = 4 and F%(2) = 16, and hence, both integral fixed points are repelling. But what about —9 and 1/3? Recall that, ‘fixed point of @ is also a fixed point for #, and since Q is an even function, we may compute Q(¢) instead of Q(=6). ‘That is, ln* — 82 Similarly, we find that e and hence, —$ and 1/¢ constitute a 2eyele for @ (see Exercise 3.3) And since F'(-4) = -49(9? - 2 ‘along with “@)-5G-9) ~¢ and 1/¢ are repelling fixed points for P. "Some authors call ¢ the golden section while stl others define it as (VB — 1/2. ‘The Inter and (VS++1)/2 are reciprocal of one another, and esgic univ apart on the real line or a particularly hid introduction he Fibonacct numbers, see chapter 11 int Ogilvy, C. Stanley and John T. Anderson. Excursions in number theory. New York: Oxford University Prose, 1968. 40 S(O) =1/2>1 = Ois repelling. S'(4em/2) = (n/2) cos(-tx/2) = 0 => -&n/2 ave superattracting 1g) Sz) fixS pers 2-2-0 = 227-2 250 o r=4y2 ‘Therefore, fix F = {0,43} P(e) = 322-1. FQ) =-1 = 0 is neutral. FsV2) = 3(4V2)? —1=5 = V2 are repelling, 1i) A(z) = arctan fix A = {0} A(z) = 1/01 + 27). A(Q)=1 = Ois neutral. e ifz 1/2 wre={ Fo. HEIR Setting each piece of this two part function equal to 2 yields fix T = {0,2/3}. Also, meal't. #E2 -l ifr<2 and therefore, (F®)(0) = F"(0)-F*(1) that the orbit of 0 is neutral. 2e) A(z) = —4 arctan(z +1) Since A(0) = —1 and A(—1) = 0, we see that {—1,0} C pery A. ‘The reader may verify that (-1)-(-1) = 1. This implies ii: 4 Aa) m(1 +(e +1)’ and (A?)'(0) = 4'(0) -A'(—1) = (-2/x) - (—4/n) = 8/x? <1. ‘This implies that 0 is an attracting pe sl wecas orey={5 HESge ah i e565 Ae Vande ie pcg ing vaya tess reo={) ife>35 point of period 2, from which it follows that (#*)'(0) = F*(0)-F'(1)-F"(2)-F'(3)-"(4) = 1:1:1-1-2=2, Thus, 0 is a repelling periodic point of period 5. 3. Suppose zo lies on a eycle of prime period 1 for the doubling function D. Evaluate (D*(x9). Is this eycle attracting or repelling? Recall the definition of the doubling map given at the end of Chapter 3: D(x) = 22 mod 1 af if0 0, itamst be true that D* (zo) # 1/2 since each such point ix eventually fae Indeed, the reader is encouraged to write down an expression for TD.43 CHAPTER 5. FIXED AND PERIODIC POINTS Bxercise 4 44 24 = D¥(20) for k= 0,1,2,....n— 1. Then (D"Y(z0) D'(x0) - D'(z1) BRB d =2 + D(tn-1) which is greater than 1 for alln > 0. Hence, all periodic points ate repelling for the doubling map! 4. Bach of the following functions has « neutral fixed point. Find this fixed point and, using graphical analysis with an accurate graph, determine if it is weakly attracting, weakly repelling, or neither. 4a) F(z) = 242" Either by inspection of the graph of F (see Figure 5.2a), or by solving the equation F(x) = 2 for x, we see that fix F = {0}. Since (0) = 1 (that is, the graph of F is tangent to the diagonal at 2 = 0) this fixed point must be neutral. Moreover, graphical analysis suggests that the fixed point is weakly attracting on the left. and weakly repelling on (0) Fa)=etet ©) Bee) =e Figure 5.2: Examples of neutral fixed points which are weakly attracting on the left and weakly repelling on the right, the right. This is a simple consequence of the fact that the graph is concave up at the origin (since F"() = 2 > 0, by the way) We remark that —1 is eventually fixed after one iteration, and that for all « in the closed interval [-1,0], F(z) — 0 as. n — 00.4 Also, for 2 <1, P(z) > 0, and therefore, F"(F(z)) —+ 00 a8 n —» 0 since is wealdy repelling on the right. 4b) F(z) = 1/2 (see Figure 5.3) ‘Multiplying both sides of Wrz by 2 and solving, we find that fix F = {1}. Also, since F@=s, wwe have that (41) = —1, and so both fixed points are neutral. But they are neither weakly attracting nor weakly repelling. ‘This is one of those atypical eases where every point is periodic (in this case with “But convergence i slow alnce the fixed poiut is neutral45 CHAPTER 5. FIXED AND PERIODIC POINTS Figure 5.8: The graph of F(x) = 1/2, a totally periodic funetion period 2). Such a map is called totally periodic, and the reader is encouraged to construct other examples of such functions. 4e) E(2) =e! (Hint: ‘The fixed point is at 2 = 1.) Although it’s easy to check that 1 is indeed fixed by B, it’s mot 0 easy to derive this result from first principles, that is, by solving B(2) = 2 for z. But we won’t let that stop us from analysing this fixed point! Locally, in a small neighborhood of 1, this problem is reminiscent of Exercise 4a where the neutral point was found to be weakly attracting ‘on the left and weakly repelling on the right. Observe that the graph of E is concave up at the fixed point (see Figure 5.2) since (1) = 1 > 0, We remark that being an exponential function, B is its own derivative Let's take a detour. The next two problems require detailed knowledge of the sine and tangent functions which is needed to determine whether the origin is weakly attracting or weakly repelling. Specifically, the trigono- metric inequalities sind <0 (63) since the base of the triangle has length 1. Now, the area of the wedge i a certain proportion of the (otal area of the unit disk, and since a circle sweeps out 2x radians, this proportion must be 0/27. ‘Therefore, the arca of the wedge is oe wr=R (64) ince the area of the unit disk is x square units. Comparing areas (5.3) and "These results can be found in most calculus texts, See, for example, Exercises 10-14 ‘on page A24 of: Anton, Howard. Calculus. New York: John Wiley and Sons, 1960, ‘Or see the lemma on pages 145-146 of: Loomis, Lynn Hl. Calcufus. Reading, MA: ‘Addison-Wesley, 1982aT CHAPTER 5. FIXED AND PERIODIC POINTS (9.4), we have sind 8 2 *2 which implies that 0<0. This result holds forall @ in the first quadrant, and may even be extended to acbitrary 0 > 0. (Exercise: Show sind <@ for 1/2 < 8 < x, for instance.) At this point, we've shown half of inequality (5.1). For the rest, observe that the area of AOP'Q is tand aT since the base of the triangle has length 1. ‘This area cleatly exceeds that of the wedge, and so tand 2 72 which implies that tand > 0. We remark that this result can not be extended to arbitrary @ > 0. This completes the detour, 4d) S(2) From the graph of S (see Figure 5.58), we see that fix = (0). But once again it’s difficult (if not impossible!) to show this by solving the ‘equation S(2) = x for 2. At any rate, we see from the graph of that sing < x for all z > 0 (sce preceding derivation) and that sinz > = for all x <0. Thus the origin is attracting, but only weakly so since sO Anticipating Exercise 8, we illustrate an alternative, more mechan ‘al approach to this problem. Observe that $*(x) = coax, 52) =S(z), and S(z) = —S%(z). ‘Thus, S'(0) = 1, $"(0) = 0, and 50) = -1 < 0. By Exercise 8, we conclude that the origin is weakly attracting. 40) T(z) = tane We know that 0 € fix? si neutral since 1”(0} 0,° and this fixed point is indeed it weakly attracting or weakly "Observe that has an infinite number of Fixed points since it's r-periodte Exercise 4 48 (0) T(@) = tans (Fe) arte (@ Fe Figure 5.5: Examples of neutral fixed points which are also inflection points. Such points are either weakly repelling or weakly attracting,49 CHAPTER 5. FIXED AND PERIODIC POINTS repelling? From the graph depicted in Figure 5.5b, it appears that tanz > z for 0 0. ‘The origin is therefore weakly repelling by Exercise 7. 4) F(z) =2+29 (see Figure 6.5) ‘This is the canonical example of a map with a weakly repelling fixed point, First of all, note that 0 is fixed by F'. Indeed, fix F = {0} Since /*(2) = 14 32%, we see that (0) = 1 and so 0 is neutral. The fact that 0 is weakly repelling follows from graphical analysis and the fact that the graph of F lies below the diagonal for negative 2 and above tite diagonal for positive 2. ‘This observation is verified using the results of Exercise 7; that is, 0 is weakly repelling since F™(0) = 0 and F"(0) = 6 > 0. We remark that the graph of F has the same basie shape as the graph in Exercise 4e 4g) F2)=2—2 (see Figure 5.54) Likewise, this is the canonical example of a map with a weakly attract- ing fixed point. Again, 0 is the only fixed point of F with F(0) = 1 and (0) = 0. But this time the graph of F lies ebove the diagonal to the left of the origin and below the diagonal to thie right. ‘This is because F'™"(0) = 6 < 0 (see Exercise 4d for a similar situation) and ‘graphical analysis confirms that the origin is weakly attracting in this case. Note that J has local extrema at x = £1 (verify this) which do not effect the local dynamics about the fixed point. 4h) F(a) = 2428 Bxercise 4 (a) Fle) @) Fe) = (6) Ble) = ~00* (@) Infe— a] Figure 5.6: Neutral fixed points with a derivative of ~1. 50BL CHAPTER 6. FIXED AND PERIODIC POINTS In this case, F has three fixed points, only one of which is neutral: = 2 2r = (27-2)=0 250 o e=svi, Since F(z) = Sx" — 1, we see that P'(0) = 1 while F% So, 0 is neutral and the other two fixed points are repelli For the purpose of comparison, the graphs of F and F? have been superimposed in Figure 5.6a. Observe that the origin is also a neutral fixed point for F#, but with (F2)(0) = I. ‘This i verified by computing Piz} (-2+25)+(-2+2°)* = (2-25) + (2 — 327 + 32 24) 830% +329 20% +2, from which it follows that (F)(x) = 928 — 212° + 1524 — 62? +1 Using the chain rule, we may check this result as follows: (Fo F)(x) = (F’o F(z) F(z) = (8(-2 + 2°)? — 1)(82? - 1) = 925 212° + 1524-62741 Vv Clearly, F2(0) = 0 and (F2)(0) = 1. Moreover, the origin is an inflection point for F? since (F2)"(Q) = 0 and (F2)"(0) = —12 £0. Since the graph of F? lies above the diagonal to the left of 0 and below the diagonal to the right of 0, the origin must be weakly attracting under iteration of F?. More importantly, p is weakly attracting under iteration of F. This is « consequence of Proposition 5.1 Let p be a fired point for F. Then p is « fied point for F*, and p is neutral for F if and only if p is neutral for F?. Proof: Suppose p is a fixed point for . Then F%(p) = FCF()) = F(p) =p Exercise 4 52 and #0 p € fix F®, Now, (PY) = FCF) FP) =O, and s0 |F"(p)| = 1 if and only if (F2)(p) = 1. ‘Thus p is neutral for F if and only if p is neutral for F?. o IU’s important to realize that (J#)(p) is always equal to one, and never minus one. We will ake advantage ofthis fact repeatedly in the sequel. We remark that Proposition 5.1 is easily extended to include the at- tracting and repelling cases. Also note that the result does not say that p is a neutral fixed point for F if and only if it’s a neutral fixed point for ¥? (can you give a counterexample’). 4i) F(z) =-2— 2° (c0e Figure 5.6b) ‘The analysis is exactly the same as in the previous exercise, ‘The trick jo to compute and analyze the dynamica of P? which in this case shows the origin to be weakly repelling. 4j) Ble) = ce* —(Ifint: ‘The fixed point is at # = —1. Examine in tail the graph of B? near 2 = —1 using higher derivatives of E?.) In Figure 5.6c, the graphs of B and #® have been plotted on the same set of coordinate axes for comparison. Note that the lone fixed point of F is also a fixed point for £* and that (E*)'(—1) = 1. These propetties, readily discernible in the figure, are guaranteed by Propo- sition 5.1: if p € fix B, then p € fix £, and p is neutral for B if and only if it’s neutral for B® Also, it appears that —1 is an inflection point for B2, and so we expect (E?)(-1) to vanish. Indeed it does, as will be verified below. A similar result will be shown true in general in Exercise 9 which says in effect that if p is a neutral fixed point for B with slope —1, then p is likewise a neutral fixed point for £* but with slope +1 and p is an inflection point for Z?. This is great since we already know how to analyze maps with these properties (see Exercises 4d, 4e, 4f, and 4g). An analysis of B? follows. It can be shown that (E°Y(2) = Ble) - (2)53 CHAPTER 5. FIXED AND PERIODIC POINTS by an application of the chain rule. Similarly, (B)"(2) = E(x) - B?(x) - (14 Elz) and (BY"@) = Ble) -B%(e) (EG) + (+ BC@))). Using these formulas, we see that (2)(—1) = 1 and (£2)"-1) = 0 which agrees with earlier observations. Also, (E®)"(—1) = -1 <0 which implies that —1 is weakly attracting for £®, and henee for E. A simple computer experiment will bear this out, 4k) (2) = ine — We may write ue) {imeo8 ifz>1 Int—2) ife<1 but note that L'(z) = 1/(#~ 1) for all x # 1. The graph of L (sce 5.64) clearly shows that fixL = {0}, and this fixed point is 1. In other words, L'behaves like 2 = —a in the vicinity of the origin. Points whose orbits come close to 0 (that is, points which are nearly eventually fixed) alternate about 0 for some time before crossing the vertical asymptote at 2 = 1? That is, the origin is weakly repelling ‘These claims may be verified using the results of Fxercise 9. Note that "(2) = ~(e— 1)? and "(z) = 22-1), and 90 £"(0) = —1 while 1"(0) Since ~21/"(0)—3[L"(0)]? = 1 > 0, it follows that the origin is weakly repelling, ‘We remark that the presence of the vertical asymptote at z = 1 causes some orbits to fluctuate wildly. Any point nearly on the backward otbit of 1 is such a point. Compute the orbits of I 2: for very small « for example. Computer experiments indicate that £7 has fixed points at approxi- mately 1.13022 and —2.06388, but this 2-cycle is repelling since (1.13022) - L'(—2.06388) = —2.5. Note also that L? has vertical asymptotes at 1, Phe reader should compute the orbit of 89, for instance, Bxercise 7 a (9) FU) = FG) = 1, Pp) > 0 Figure 5.7: Two cases of a neutral fixed point. 5. Suppose that F has a neutral fixed point at p with F"(p) = 1. Suppose also that F"%(p) > 0. What can you say about p: is p weakly attracting, weakly repelling, or neither? Use graphical analysis and the concavity of he graph of F near p to support your answer. Since F"(p) > 0, the graph of F is concave up at the neutral point p (see Figure 5.7a). Graphical analysis clearly shows that p is weakly attracting con the loft and weakly repelling on the right. 6. Repeat Exercise 5, but this time assume that F“(p) < 0. When P"(p) <0, the graph of F is concave down at the neutral point p (cee Figure 5.7b). In this case, p is weakly repelling on the left and weakly attracting on the right. 7. Suppose that F has a neutral fixed point at p with Fp) = 1 and FP") = 0. Suppose also that F”"(p) > 0. Use graphical analysis and the coneavity of the graph of F near p to show that p is weakly repelling. When F!(p) =0, we know from calculus that p will be an inflection point for F if F"(p) ¢ 0. Now suppose F"(z) is negative to the left of p, and positive to the right of p. Then the graph of Fis concave down to the left, and concave up to the right (see Figure 5.8a). Graphical analysis shows that55 CHAPTER 5. FIXED AND PERIODIC POINTS (a) PQ) =9,F'@) = 1,F"() = 0, and (6) FP) PMD) > 0 FMp) 0 8. Repeat Exercise 7, but this time assume that F’”(p) <0. Show that p is weakly attracting. Tin this case, we suppose that F"x) is postive to the left of p, and negative to the right of p, s0 that the graph of F is concave up to the left and concave down to the right (see Figure 5.8b). Arguing as above, it follows that F"(p) <0 9. Combine the results of Exercises 5-8 to state a Neutral Fixed Point ‘Theorem, ‘The four basic cases are illustrated in below. ‘igure 5.9 for p = 0 and summarized ‘Theorem 5,2 Let p be a newtral fized point for F with F"(p) = Case 1: Suppose F"(p) #0. If F"p) <0 (resp. F"(p) > 0), then p is Exercise 9 56 a7 ©) FW) =OFW)= 1.4) >0 (bY FCO) =0,FYO) = LP") <0 aa (©) FO) =0,F(0) = 1,P(0) m0) >0 and (4) F(0) = 0,F"(0) = 1, P"(0) = 0, and FO) <0 Figure 5.9: The Four Canonical Forms of Neutral Fixed Points.a7 CHAPTER 5. FIXED AND PERIODIC POINTS weakly attracting (resp. weakly repelling) on the right and weakly repelling (resp. weakly attracting) on the left Case 2: Suppose F"(p) = 0. If F'"(p) <0 (resp. F"(p) > 0), then p is ‘weakly attracting (resp. weakly repelling). But what if F"(p) = —1? As we saw in Exercises 4h-4k, the trick is to apply ‘Theorem 5.2 to F®. First of all, by Proposition 5.1, a neutral fixed point for F i also a neutral fixed point for F?, but with (#){(p) = 1. Furthermore, using the chain and product rules for derivatives, we find that (FY) = PFC) Pel + PF) -P"2). Evaluating thi derivative at p, we obtain PEP) LEP + FF) - PP) "(n) — PM) YO) and so case 2 of Theorem 5.2 applies. What remains is the somewhat tedious computation of (F2)'"(p}. The more industrious reader will be inclined to verify that, (ery) = FUE (2) [GP + BE" (@)) - Fe) F(a) + F(R (@)) F(a) from which it follows that (Fy) = -2F"() HEP. ‘When this quantity is negative, the neutral fixed point p is weakly attract- ing for F?, and hence for F. Let's summarize this result in the following Theorem 5.3 Let p be a neutral fized point for F with F"(p) = —1. If the quantity ~2F"(p) — 3[F"(p)} is negative (resp. positive), then p is weakly altracting (resp. weakly repelling). Whe ter) ‘he some as saying F has negative Schwarzian derivative at p. (See Chap- Exercise 9 58 Unfortunately, there are still cases of neutral fixed points that have yet to be considered. What if F"(p) = 1 and F"(p) = F"(p) = 0, for instance? Under these conditions, Theorem 5.2 does not apply and we must continue evaluating higher derivatives at p until one of them is nonzero. Even this will not always work, however, sinee there ate maps having a neutral fixed point satisfying F'(p) = 1, and for which F%(p) = 0 for all n > 1. ‘The identity map is one such (trivial) example—can you find others?Bxercise 1 60 Chapter 6 Bifurcations | (0 Some tcl any mars (0) scatindngr Exercises Figure 6.1: Bifureation behavior in the quadratic family Fy(2) = 2427+ 1: Racket ase nai aon tex Wi a RE pole petreprsrsirvienen. eppeehpensiatatcboratatichar nee ao ee tus to tent this bifurcation ae either «sade mode or pelodcdoubling bifurcation, or neither of these. In each case, sketch the phase portrait for which allows us to determine the \ values for which Fy has an attract- typical parameter values below, at, and above the bifurcation value. ing fixed point: Ja) Fy(z)= 242740, A=0 -1< Fy(-v=¥) <1 ‘To find the fixed points of F’, we must solve & -1<1-2vNe1 abet tA 2 2e-2-K<0 loco thay 3 1>v=R>0 or 2. This yes y e=4v—. > 1>-A>0 nae . S$ H1 0, one fixed point for 0, and € pair of fixed points for A < 0 (sce Figure 6.la). Clearly, A= 0is a We conclude from this that ~/=N is attracting for 1 < 4 <0. On bifureation point for F, and note that 2 = 0 when \= 0. Using the the other hand, +V—A is always repelling since F{(4V—A) > 1 for fact that P(e) = 2e-+1 we se that this point iain fact neaddlonode all <0. See Figure 6.1b, bifurcation since F4(0) = 1. Observe that F{ is independent of IDA@)ee+e 4A, Xe-1 But we can say more. First of al, its easily shown that ~(1-+ =X) : i Z ean hall nee ‘This problem is a continuation of Exercise 1a (see Figure 6.1), Recall is minal fixed. Secondly, if we evaluate F{ at the negative fixed that —V=X is attracting for 1 < A <0 and that Fy experiences a Point, we get saddle-node bifureation at A= 0. Another bifureation oceurs at the BY(-V=R) = 2(-V=A) +1 left-hand endpoint of this interval. Indeed, when 4 = —1, we haveo CHAPTER 6, BIFURCATIONS Exercise 1 62 (2) Some typical family members and (b) Bifurcation diagram. their second iterates. Figure 6.2: Bifurcation behavior in the cubie map F(z) 1 with PLy(-) =-1 a neutral fixed point at 2 = —/—(-1) Hence, F undergoes a period-doubling bifurcation at 1c) Fy(z) = Ae t2%, AS Set Fy(2) equal to x and solve: Mrtarae > Q-Dete? = (Q-N+)2=0 sp asdvI=X or 2=0. When A> 1, there is a single real fixed point at the origin, and for A <1 there are three fixed points (see Figure 6.2a). From F{(z) = A+ 822, we compute F{(0) = 2. Hence, the origin is attracting for =1. Once. ‘Typical members of the sine family $,(2) = pesin 2 Figure 6. and s0 the fixed points 4:VT=% ave attracting when -1<3-2<1 4c Ac? 3 2>A>1, but note that +VT=T% is complex in this range. Soe Figure 6.2b, ‘These calculations suggest that F, undergoes bifurcations at A= —1, 1, and 2, When \ = —1, we have already shown that there are three fixed points, and for this parameter value we find F'(0) = —1 and FL4(4V2) = 5. Hence, there is a period-doubling bifurcation at the origin when = 1 and the other two fixed points are repelling. 1d) Fa(a) = Az +29, A=1 ‘This problem is a continuation of Exercise le. When 4 = 1, x = 0 is fixed, and we've already shown that the origin is attracting when =1 1. Sy(z) = sinz, p=—1 (see Figure 6.3b) ince $!4(0) bifureation at js attracting 2cycle ‘The reader may wonder if there other bifurcation points for Sj, and if s0, what are they? We begin to answer this question below. 1 =1, it appears that Sj, undergoes a period-doubling We remark that for x < 1, Sy has an Since bifureations occur at neutral fixed points, what we need to do is solve the equations ping = 2 (6.1) and cos s = 1 (62) simultaneously. Now, if we divide (6.1) by (6.2), we get tang =e for. # 0. (The tangent function was cursorily examined earlier in Exercise 5.4e.) In other words, the bifureation points of S, are the fixed points and 2-cycles of x tanz.! th) Ex(z)=Mef = 1), A=—1 Note that (0) = 0 and 90 0 € fix Ex, Also note that BX(2) = Aet = Ex(z) +A. Thus, £4(0) = A. ‘Therefore, the origin is attracting for | < 1 and repelling for [A] > 1. When A = =I, By undergoes a petiod-doubling bifurcation since _,(0) = ~1. See Figure 6. i) A(z) =MeP= 1), A=T ‘We have from the previous problem that, B3(z) = Ex(2)+A. It follows that. (0) = I which shows there’s a saddle-node bifurcation at \ = 1. See Figure 6.4 for the graph of Ey Ij) H(z) = 24 02%, 0 "Recall that tangent is an odd function, and that the 2- 0, H"(0) > 0, and the origin is weakly repelling on the right. and weakly attracting on the left. But when ¢ = 0, H1"(0) = 0, and we look to case 2 of ‘Theorem 5.2. Unfortunately, H?"(2) is identically zero and so the theorem does not apply. Observe, liowever, that Ho is the identity map, which is totally periodic. In summary, this family of maps experiences no bifurcations whatso- ever, and provides @ good example of why the precise definitions given in Chapter 6 of the text are necessary. Ik) Ps) =eter2+e8, c= 0 Titst of all, when ¢ = 0, Fy is identical tothe map in Exercise 14 with d= 1. (See Fy and F? in Figure 6.22.) But a generic member of this amily of functions has two fixed points (see Figures 6.6b-c) smce etotprae = ex? +29=0 = (ete)? =0 Seance o 2=0. 14 2ex + 32”, and so F{(0) = 1 regardless of ¢. Hence, fixed point fails to undergo a bifurcation which ean be seen in Figure 6.6d. Note that. F{(—c) = 1+. which is strictly greater than one for all ¢ £0. Thus, ~e is repelling for all ¢ (even ¢ = 0 which is weakly repelling). We also remark that F.(—2) = —F,(z), a most curious property. ‘The next four exercises apply to the family Q-(z) = 2? +c. 2. Verify the formulas for the fixed points px and the 2-cycle gs given in the text. Recall that 1a vine pan 24 Exercise 2 66 4 tld (2) F and it second iterate, (6) Poa and its second iterate Figure 6.6: Representatives of the family F(z) = (2) Bifurcation diagram for ¢< 0, texte,67 CHAPTER 6. BIFURCATIONS and note that these points are real when 1 —4e > 0, that is, when ¢ < 1/4. Now, Qelo.) = Qel(1 + VI=Fe)/2) 1+ VI=4)4/4 +6 14+2vI—4e+1—de)/A+e (2+ 271 — de — 4c + de)/4 14 Vint Ps ‘We leave it to the reader to show that p_ is fixed. Now, let’s verify that ew aceel 2 on constitute a 2-eyele for Qe: Qela4) = Qe((—1 + V=4e=3)/2) (14 VIA +e (1 = 2V=Fe=3 — 40-3) +e (2 2VAe=3 — 4e-+ 4e)/4 (-1- v=4e=3)/2 =a ‘The reader may show that. Qe(q- We remark that this 2-cycle exi as well. Thus, {q4,4-} C perz Qe. for —de ~ 3 > 0, that is, for e< 3/4. 3. Prove that the cycle of period 2 given by ¢4 is attracting for —5/4 < e<-3/A First ofall, note that Q2(2) = 22. We have that (Q2) (a) = Relay) Qi) 1+ Ve 8) «(1 V=te=3) = (-4e=3) +4 (6.3) which basically solves this and the following two exercises. ‘To find the parameter values for which this 2-cyele is attracting, we must solve -l B Qet(1-a)z=0 S PsQ-1/A or 2=0. We may conclude from this that wn Asa check, let’s compute69 CHAPTER 6. BIFURCATIONS =a-9(!=2=2) Now, ‘Thus, (A — 1)/2 is attracting for -1<2-d<1 3 Behe = 1 3. Exercise 14 70 Figure 6.7: Bifurcation diagram for the logistic family. Note that which suggests a 2-cycle may be lurking in the shadows. 14. Compute an explicit formula for the periodic points of period 2 for Pa. First compute the second iterate of Fy’ FR (2) = AQa(1 ~ 2))(1 Az —2)) (Az — Aa?)(1— Ae + Az?) = AQe — 22? + N29 — de? + M29 — 2a) (Aa — (A+ A2)2? + 24225 — 224) = Me — (1 + A)a? + 20829 — a2 "find the fixed points of Ff (ie. the period 2 points of Fy), set FE(2) equal to, rearrange terms, and got (3? = 1) = (1+ Aja? 420%29 — 84 = (65) ‘This dUtidegree polynomial may be factored into a pair of quadratics. One of these factors must be (A~1)z— A2? since the fixed points of Fy are also fixed points of FZ (see Equation 6.4). Long dividing out this 2nd-degroea CHAPTER 6. BIFURCATIONS Exercise 16 polynomial from (6.5), we find that, (22 = He — °C 4 a)? + 20824 — at Q=De— de 2? + e+ (+1) and 80 [A= te A212? — A+ e+ 04 Y)=0. This new factor may be solved using the quadratie formula: Da _ AGH) 4 aAVOF IFFT ~ PU _ ++ VOFIO-3 ~ 2 Hence, this 2-eyele exists when > 3 or A-< —1 which agrees with the resulls of Exercise 8.Chapter 7 The Quadratic Family Exercises 1. List the intervals which are removed in the third and fourth stages of Une cousteuction of tte Cantor micldle-thirds set. Step 1: (1/3,2/3) Step 2: (1/9,2/9)U (7/9,8/9) Step 3: (1/27, 2/27) U (7/27, 8/27) U (19/27, 20/27) U (25/27, 26,27) Step 4: (1/81, 2/81)U(7/81, 8/81)U(19/81, 20/81)U(25/81, 26/81)U(59/81,56/81)U (61/81, 62/81) U (73/81, 74/81) U (79/81, 80/81) 2, Compute the sum of the lengths of all the intervals which are removed from the interval (0, I] in the construction of the Cantor middle-thirds set. From Exercise 1, it should be clear that 2"~! open intervals are removed at the nth stage of the construction of the Cantor midale-thirds set, each having width 1/3". ‘The combined length of these intervals is 1 1egeaeg but Exercise 6 a“ In other words, construction of the Cantor middle-thirds st isa subtractive process which removes an infinite number of open intervals with combined Tength equal to that of the original unit interval. Apparently, little is left behind at the conclusion of this process. ..or soit seems. In the next five exercises,’ find the rational numbers whose ternary expan- sion is given by: in almost all cases, the techuique is to group succescive pais oF tiplts of terms and simplify. Grouping of terms is permissible sluce a geometric series is absolutely convergent,15 CHAPTER 7. THE QUADRATIC FAMILY “7 ‘igure 7.1: The graph of T(2) = 3/2 — [Bz — 8/2) agp it ae if set + lets ay Dynamics on the Cantor middle-thirds set: ‘The following exercises deal with the function meal se f2si/2 re={ Fe ite >1/2 9. Sketch the graph of T’ and show by graphical analysis that, if x > 1 or 2 <0, then T"(2) + -00 a8 n+ 00. See Figure 7.1. Exercise 12 6 10. Find the fixed points for 7, What is the ternary expansion of these points? ‘The origin is obviously a fixed point. The rem: by solving ing fixed point is obtained S-dese > 342 > 32 Cheek: 18/4) = 3 38/4) = 3-9/4 = 19/4—-9/A= 3/4. ‘The ternary expansion of 0 is trivial, but what about 3/4? We know from the example on page 7 of the text that _ ea fl\i_ 1 om=25° (3) ; So all we need to do is multiply each ofthese quantities by 8, But multi plying a ternary expansion by 3 shifts the ternary point one place to the tight. Thus, 3/4 = 0.20. This may also be seen by weiting le7 CHAPTER 7. THE QUADRATIC FAMILY 12, Show that if x € (1/3,2/3), then T*(2) + —o0 as n = oo Note that is strictly increasing on (1/3,1/2] and strictly decreasing on [1/2,2/3). Therefore, dewcd w r(3)1 So, any way you look at it, 1 < T(z) < 3/2 when « € (1/3,1/2), and in Exercise 9 we showed that if > 1, then 7"(2) + —oo as n oo. 13. Show that if 2 € (1/9,2/9) or x € (7/9,8/9), then T™(z) > oo as to this problem requires some knowledge of T?. To find an for T(z) we begin by funding those 2 for which 7x) < 1/2. 1 2 < 1/2, then T(x) = 3x < 1/2 when # < 1/6. Similarly, if x > 1/2, then T(c) = 3— 3x < 1/2 when x > 5/6. These results suggest we compute the value of T%(z) on the following partition of the real line: 2<1/6 = M2) <1/2 = T2)=T(T(2)) =T(32) =3(82) =o: YOSe W2 TYs)=T(7(2)) =T(3~32) 3(3 32) =9-92 25/8 > M2) <2 Exercise 14 8 Figure 7.2: The second iterate of T(x) = 3/2 — [8x — 3/2) ‘Thus, we have shown that Oe 3-98 92-6 9-9 12) = See Figure 7.2 for the graph of 7?. show that, sir? = {o, sig? ifz<1/6 ifl/o<2<1/2 if 1/252 55/6 if 25/6 By the way, the reader is invited to 339 } 10 We're now ready to solve the original exercise. Note that 7? is strictly increasing on (1/9, 1/6] and strictly decreasing on [1/6,2/9). Therefore, 1ooel 9<75G 9 and 12 t If. € (1/9, 2/9), then T(z) € (1,3/2], and by Exercise 9 it follows that TM(z) + =00 a8 n+ 00.79 CHAPTER 7. THE QUADRATIC FAMILY Similar arguments show that if 2 € (7/9,8/9), then T*(z) + —o0 as n+ 14. Let P= (2 € [0,1] | T%(2) € [0,1] for alm}. Prove Cantor middle-thids set. Let z = 0.a,azas .... and partition the unit interval so that [0,1] = [0,1/3]U (1/3,2/3)U[2/3, 1}. One of the key ideas of the proof is that the value of a, alone determines which of these three intervals contains z. If a: = 0, then +» € [0,1/3} if a; = 1, then x € (1/3,2/3); and if a; = 2, then = € (2/3, 1) At first glance, the boundary points appear to be exceptions to this general rule since 1/3 = 0.10 and 2/3 = 0.12, but even these may be written as 0.02 and 0.26, respectively. Another key idea. is the important result of Exercise 12: all x in the interval (4/3, 2/8) have orbits which shoot off to co. But these are precisely those 2 whose first ternary digit is equal to 1, Indeed, the claim is that any 2 having some ternary digit equal to 1 has an orbit which escapes to —oo! ‘This is true because KK, the =f Cazares... 00 <1/2 me={ casi ages where 0 ifa=2 @=d 1 ifa=1. 2 ifa=o0 Note that this definition of Tis unambiguous at x = 1/2 which itself has ternary expansion 0.1. Now the proof that I’ = KX goes asfollows. Suppose # €[. Then the ternary expansion of 2 can not have a 1 in it, for if it did, T*(2) € (1/3, 2/3) for some and the orbit would escape to —o0. ‘Thus, «is alsoin K. Conversely, suppose 2 € K.. ‘Then the ternary expansion of 2 again has no 1s. ‘Thus, T*(x) ¢ (1/3, 2/3) for all n and so the orbit of 2 can not escape. Hence, 2 isin. 15. Suppose 2 € T’ has ternary expansion 0.a;aza3..-... What is the ternary expansion of T(x)? Be careful: there are two very different cases! By the combined results of Exercises 9 & 12, we know that either # € [0, 1/3] ot « € [2/3, 1]. If'z € [0, 1/3}, then its leading ternary digit is 0; otherwise, is 2. ‘These are the two cases that must be dealt with, Exercise 15 80 ay =0, then T(2) = Ss. As we've already seen in Bxercie 10, multiplying a ternary expansion by 3 simply shifts the ternary point one place to the 2, then 2) = O.aya20s..4 dat... where Putting all these facts together, we have that for 2 €T.Chapter 9 Symbolic Dynamics Exercises 1. List all eycles of prime period 4 for the shift map. ‘There are twelve periodic points of prime period 4: *(0001) = 3 (0010) = «*(0100) = o 1000) = (GOTT); * (OTTO) = 0° (1700) = o( TOOT) = (GOT); °(0TTT) = o°(TT10) = o( 1101) = (TOT), Note that all 2-eycles and fixed points are trivially 4-cyces Compute dfs, t] wheres 2, 8 = (0), t= (00) C1), 000} = (143) + ($45) + (Gag) + $ 5 5 Bxerciso 4 82 3. «= (10), t= O10) (100), (O10) = Note that (O10) is further away from (FOO) than (O01) because it differs from (100) in the second place while (O01) differs in the third. Both (010) and (001) differ from (700) in the frst place, and so each must be at least ‘one unit away. But how far are they from each other? (001 (070) = (5+ 3) + (aq +a) + (spa +t ape) to Thus, they are closer to each other than they are to (TOO) because they ‘agree in the fist position 4. ©= (01), €=@D a[(1011), (O107)] =83 CHAPTER 9. SYMBOLIC DYNAMICS Considering the fact that the maximum distance between any pair of strings in ¥ is two units, the above 4-cycles are surprisingly far apart. (Do you see why?) 5. Find all points in 5 whose distance from (000. ..) is exactly 1/2. ing ¢ must be in Mo, otherwise d{(000...),t] > 1 2, First of all, any such st Tt tums out that there are two strings ¢ such that d{(000...,¢] namely, (010) and (001). This result may be generalized. Let » = (a99...). There exist two strings ¢ such that afs,¢] = 1/2%, namely, (6081 -.-Sntatitny2.-.) and (S081 ---Snin418n42 +) 6. Give an example of a sequence midway between (000...) and (LI... Give asecond such example. Are there any other such points? Why or why not? One of the two points midway between (0) and (I) is (01) since d{(0T), ()] = (01), (1)] = 1, The other point is (10), and these are the only such symbol soquoiteos by virtue of the Proximity Theorem, T. Let Mo: = {8 € © | 59 = 0,1 =1) and Min = {8 €2| 80 = 1,5; 0,82 = 1}. What is the minimum distance between a point in Moy and point in Afjo:? Give an example of two sequences that are this close to each otter Let s € Mo: and ¢ € Mot. Then dfs, t] > 8/2since s and t differ in the fist two positions, For example, let s = (O11) and ¢ = (101). ‘Then als, ¢] = 3/2. More generally, suppose t = (101ata...). Then # = (OLltsty...) is exactly 3/2 units away from t. 8. What is the maximum distance between a point in Mo and a point in Mio:? Give an example of two sequences that are this far apart. ‘The maximum distance is 2 units since the strings need not agree at any position. Given any t € Myo, all strings in Moy of the form (010iat,....) are 2 units away from t. ‘The N-Shift: ‘The following seven exercises deal with the analogue of the shift map and sequence space for sequences that have more than two possible entries, the space of sequences of NV symbols. Exercise 12 84 10, Let Ey denote the space of sequences whose entries are the positive Wegers 0,1,...,N—1, and let ow be the shift map on Sy. For 6, € By, let dys, t] Prove that diy is a metric on By. ‘The function diy is nonnegative since |s; —ts| > 0 for all i, and it vanishes if and only ifs = t. So all we really need to show is symmetry and the triangle inequality. Symmetry follows since |s; ~ tj] = [ty — 4] for all numbers 5; and fj. The triangle inequality is a consequence of the triangle inequality for ordinary numbers: dy(s,t]+dy{t, ‘Thus dy is a metric and (dy, 2y) is a metric space. of sequences in Dy? stance between a pi 11. What is the maximal ‘The maximum value of [si — ti] is N —1, and therefore, the maximum distance between two sequences in Ey is So, in general, the maximum distance can be no more than the size of the alphabet.85 CHAPTER 9. SYMBOLIC DYNAMICS 12. How many fixed points does oy have? How many 2-cycles? How many cycles of prime period 2? Let ow be the shift map on Ey. Specifically, let ow: Sy —+ By with ow(s08182...) = (s18285...) Now ow has N fixed points; indeed, fixow = {(000...),(11.-.),-+-5(BbE...)} where k = N~ 1. Recall that o has two points of prime period 2 and we wonder ify has N points of prime period 2. It tums out that this is not the \ce any sequence of the form (5557) is of period 2, and there are N? such points, But V of these are fixed, and so there are N?—N = N(V—1) points of prime period 2 13, How many points in Dy are fixed by of)? For starters, how many periodic points of period 3 are there? We have that pera ow = { (545732) | sn, 1,32 € Ev } ®, Of these, N°—N = N(N?—1) are of prime period im ate fixed, IU not very difficult to see that and s0 | perg oy Bssince NV of tl Iper, ow] = N", in general. For example, on has N* 4-cycles and N*— N* = N2(N?—1) sequences of prime period 4, but it’s not clear how many points of prime period n there are in general. 14. Prove that ow: Ey — Ey is continuous. First we give a careful proof that ¢:8 — © is continuous, a result: which boils down to the following important fact: Lemma 9.1 If d{(s0s182...), (torte...) < 1/241, then al(sisees .-)s(Citats...] < 1/2". Proof: If d{(sosiso ...), (lolita...)] < 1/2"*%, them 9; agrees with ty for i € {O1,....m +1} (by the Proximity Theorem). And if s; = tt for i € (0,1,...,m+1) (and hence, for i € {1,2,...,n-+ 1}), then f(si9283-..),(¢rtata...)) < 1/2", Exercise 15 86 also by the Proximity Theorem. DB Using this result, the proof that o is continuous proceeds as follows: 1 let € > 0 be given, and 2, choose n such that 1/2" < ¢; 3. let 6 = 1/2", and 4. suppose df(sosisz....),(totita.-.)) < 5; 5. then dl(sisrsa...),(titata-..)) $1/2" <6. Next we need a generalization of the Proximity Theorem. Let s,¢ € Ey and suppose 5; = t; for i € {0,1,...,n}, then ,8)= < tt =w-yie _a “ie ‘The converse is also true: if d[s,t] < 1/N™, then sj = ts for i € {0,1,-..yn} We remark, however, that it is not the case that s; = ty for i € {0,1,-.-)n} if and only if dls,t] < 1/N", and the reader is encouraged to find’ an appropriate counterexample (sce Exercise 5 for a hint). To show that ow is continuous, let ¢ > 0 be given and choose n so that 1/N" < ¢. Now suppose 6 = 1/N"* and d[s,t] <6 ‘Then 3; = 4; for 4€{0,1,...5041}, and dlow(s),on(t)] $ 1/N" 0 since &4(s,t) > 0 by definition in all three cases. But is dels, ¢] = 0 if and only if's = t? ‘The answer is “yes” in the case of (9.1) and (9.2), and “no” when (s,t) = 6» +t» mod N (cince 6-((0),()) = 0, for instance) Both (9.1) and (9.2) are symmetric, that is, ds[s,€] = da[tys], but what of the elusive triangle inequality? Consider the situation in (9.1) where we must show that 5,(6, t)+4e(t,ut) > 6,(s,u). ‘There are four cases to check: fu(ost) elton) Baloo) 0 0 0 0 1 1 1 0 il 1 1 Oorl In all cases, 64(8,) + &c(t,) > 6t(s,u) and s0 ds satisfies the triangle inequality. Likewise, enumeration will also show that (9.2) gives rise to @ metric space. 16. What is the maximum distance between two points in By when the previous metric dy is used? Since 6:(s, t) given in (9.1) is at most 1, dys, t} ean be at most N/(N — 1) because dale, = SoD * op Exercise 18 88 We leave it to the reader to determine the maximum distance between two points in By for 6:(8,t) in (9.2) 18. Bach of the following defines a function on the space of sequences © In each case, decide if the given function is continuous. If s0, prove it. If not, explain why. 18a) F(s08162...) = (Ososis2. Let s = (sosis2...) and t F(t) = (Ototits..), and al (s), F(e)] = ‘This suggests we prove F continuous as follows: Let ¢ > 0 be given, and choose 7 such that 1/2" < ¢ Now let § = 1/2"! and suppose dis, t] <6. Then 1 ese ars), Fee) = Sale, < 36 and so F is continuous. 18b) G(s081 82...) = (080081082...) We proceed as in Bxercise 18a, In the present case, d{G(s), G(t)] is strictly less than d{s,t]/2 however, sinee89 CHAPTER 9. SYMBOLIC DYNAMICS ! Exercise 18 90 Now, the claim is that for s, t € 2, Fabs. S$ d[H(s), H(®)] < 2als, (0.8) IE you believe this, then the continuity of Hf follows almost immedi- ately. As usual, let ¢ > 0 be given, and choose n so that 1/2" < ¢ ‘Then let § = 1/2" and suppose d{s,t] < 6. But d[F(s), H(t)] < 2dfs,t] by (9.8), and moreover, 2dfs, t] < 26= 1/2" 0 be given and choose n s0 that 1/2" < ¢. Now let /2?* and suppose dfs, t] < 6. Then by the Proximity ‘Theorem, 4 for i= 0,1,...,2n, and 50, das), 200) = So Pastel mt e3t2 * ee pee Tpa1 CHAPTER 9. SYMBOLIC DYNAMICS Exercise 19 ‘Thus L is continuous. 18g) M(s081 82 ...) = (0810810081000 ««.) ‘This exercise parallels the previous one, except this time we must choose 6 to be very, very small. Specifically, we may choose § = 1/2!" 80 that s and ¢ agree on the first 10" + 1 entries. 19. Define a different distance function don ¥ by ds, t] = 1/(k-+1) where 1 is the least index for which sp # te and d'[s,s] =0. Is da metvic? Yes, d’ is a metric. Observe that d'fs,t] > 0 for all s and t, and that. @'[s,t] = 0 if and only ifs = t. That @[s,t] = a'[t,s] is also obvious from the definition. But does a’ satisfy the triangle inequality? Let s, t, and u be symbol sequences in 33, and let £y = mini €N | a # t} and ky = min{ FEN |G fs}. A key result is as, u) bat clearly, 1 1 1 inin(li, fa) * min{Fy Fay * inn(Ey, Fa) Ll ath and so the triangle inequality is satisfied 92Chapter 10 Chaos Exercises For each of the following sets, decide whether or not the set is dense in (0,4) 1. S; is the set of all real numbers in [0,1] except those of the form 1/2" for n= 1,2,3,.... Yes, Si is dense in (0, 1]. One way to show this is to first let w = 1/2” for some fixed positive integer n, and then produce a sequence zx in 51 such that ry + was k +00 We use a biseetion technique on the interval (w,2u}. Begin by letting 0 = 2w = 1/2" and forming the sequence wea 2 a= for k > 0. For example, whey 1/24 1/20 a= 2th = ey = WEE Vr ape 7 2 we, 1/2r 4 spe? z Exercise 2 4 and in general, it appears that noth °= a which the reader is asked to show by induction. Note carefully that zp € 5 for k > 0. We claim that 21 —+ w as k = oo. ‘To see this, consider the function wee 2 where w is any real number. Note that Fy is a linear map with attracting fixed point w. Moreover, its basin of attraction is the whole real ine! In summary, let w= 1/2" ¢ S,, eo = 2w, and Fo(2) = (w +2)/2. Then F&(zo) + w as k — co, and F(z) € 5; for all positive integers k. This proves that S; is dense in (0,1) Fale 2, Sp is the set of all rationals in [0, 1] of the form p/2" where p and n are natural numbers. Note that S2 contains all multiples of 1/2, all multiples of 1/4, ete. In fact, # € Sp if and only if = has a terminating binary expansion. Assuming this o be true for the moment, we may easily show that $2 is dense in (0, 1] as follows: Thet w = O.dbabg... be an arbitrary point in [0,1]. If w terminates, then wwe are done, #0 suppose it does not. ‘Then the sequence Ob, O.brba, O.bybaba, obviously converges to w, and each element of this sequence is in $2. qed. Using a bisection technique, we now exhibit such a sequence in Sa converg- ing to w. Take the unit interval, divide it in half, and determine which half contains w. Discard the half which does not. Halve the remaining interval, ‘and again ask which half contains w. Continue this halving process, each time throwing away the half interval which does not. contain w. ‘This bie nary search technique, as it’s called, captures w to any degree of precision cone cares to specify. A corresponding. algorithm is given in Figure 10.1. The variables J, and re are the left-hand and right-hand endpoints of the subintervals, respectively, fe say that wis a globally attracting fixed point for F95 CHAPTER 10. CHAOS Loop mg = (lk + re)/2 it wm, then be = me end if ksk+1 end loop Figure 10.1: A binary search algorithm for w. and my is the computed midpoint, We'll describe the by in a moment, but first observe the following facts: Thoth t, and mate in 2, and hence, m, € Se (in fact, m, = 1/2); arguing inductively, if ly and ry ate contained in S2, then so is m™., and hence, all and ry are contained in $2; 2. since w gS, w my for all k; 3. my > was k +00. ‘The binary search algorithm in Figure 10.1 can also be thought of as a binary tree with the elements of S at the nodes, and Os and Is decorating the edges. See Figure 10.2 In this tee, a movement to the left traverses an edge marked with a0, while 1a motion to the right picks up a 1. ‘These bits correspond to the binary expansion of w, and are precisely the by given in the algorithm. In fact, the concatenation of all these b, corresponds to one of the terminating binary expansions mentioned earlier. 3. Sp is the Cantor middle-thirds set. ‘The Cantor set K can not be dense. There can’t possibly be a sequence in K converging to 1/2, for example, since the interval (1/3, 2/3) was removed al the first stage of the Cantor set construction. Exercise 6 96 ie, A XN, NN; oO YN A A Figure 10.2: A binary tree of the elements in S. 4. Sx is the complement of the Cantor middle: thirds set. Recall that the ternary expansion of z € K has no Is. Any point which does have a 1 in its ternary expansion is in Ks complement. Moreover, any finite string can be prepended to such a point, and the result. is sit! in K's complement. Thus the complement of K is dense in [0,1] since there are (uincountably many) points in K's complement arbitrarily elose to any point 2 in the Cantor set. Just prepend the first n + 1 bits of 2 to your favorite point in the complement of K 5. Ss is the complement of any subset of 0,1] which has countably many ‘elements, Note that the set 5; in Exercise 1 is a special case of Ss. Yes, the unit interval remains dense even after removing a countably infinite number of points. ‘The following argument depends on the fact that open intervals are un- countable sets.? Let 2 be any point in the complement of Ss, and let ¢ > 0 be given. Now let V = N(z) be any cneighborhood of z. Then N 115s is nonempty since WN is an open interval and hence, uncountable, Hence, S is dense in (0,1) For each of the follo Give reasons. 6. T= {(s08182.-.) | 54 = 9}. ig sets, decide whether or not the set is dense in ©.97 CHAPTER 10. CHAOS Any point s ¢ 7} is of the form (3081828318885 ...). The point in 7}, closest, to s is (sos1s2s90se56...), and in fact, {(s051 528215580 ...), (8081828508556 ‘Therefore, 7; can not possibly be dense in ¥. 7. Tis the complement of Ty. Observe that Ty identical to the onc ((sosisa-.) | a4 = 1}. By am argument virtually Exercise 6, Tz can not possibly be dense in 5. 8. Ty ={(sosrs2-..) | the sequence ends in all 0s} ‘We will show that 7s is dense, Take an arbitraty 8 = (sos162...) in Es and construct a sequence of points sq such that, fa = (0081-44-10), Note that each s, € 73 and that 8, — s as n + oo. Thus Ty is dense in ¥. (See Fxersine 2 for a related prablem ) 9. T= {(sosie2...)| at most one of the 5; = 0) ‘This set can not possibly be dense in 3. Consider the point (001) in 74's complement. ‘There is no sequence of points in 7, converging to (001) Indeed, the closest point in Ta to (001) is (OT), and a{(o0T), (OT)] = 1/2. In fact, there is no sequence in Ty converging to any point in its complement. Let 8 ¢ 7). Then s has at least two 0s. Now suppose sy is the second of these two 0s. ‘Then each point in 7) is at least 1/2* units away from s, and ‘0 there can be no sequence in T; converging to it, 10. Ts = {(#08332....) | infinitely many of the +; ‘Take a point in Ty's complement which has but finitely many 0s. ‘Then it has tely many Is, and in fact, the tail of the sequence must be all 1s and hence ofthe form (sos1 al) for some n. But the sequence of points (e081 --.800), (#081 --.8010), (8081... 110), converges to (8081 ...SnT), and each such point is an element of Ty. ‘Thus Te is dense in E. (See Exercise 8 for a related problem.) Exercise 16 98 11. Te is the complement of Ts, As argued in Exercise 10, a point in Ts must end in all 1s. Consequently, we proceed as in Exercise 8 by constructing a sequence of points, each ending in all Is, converging to any point in ‘We remark that Ty is not the same as { (608182...) | infinitely many of the #) = 1} since there are strings in the latter which are not in Ts. ‘The string (UT) is ‘one such example, 12. Ty = {(s0tr42-..) | no two consecutive 4; = 0} In words, Tr consists of those strings in which every 0 is followed by a 1 ‘The complement of this sot consists of al strings with a consecutive pair of (s, The string (007) is one such example. Unfortunately, there is no string in Tr close to tis string—in fact, the element in Ty closest to (007) is (07), and df(O01), (01)] = 1/2. Therefore, Tr is not dense. 13. Ts is the complement of Ty. ‘The complement of Ty is dense in Y. As mentioned in Exercise 12, Ts is the set of strings containing a consecutive pair of Os. Now take any point in B and construct a sequence in Ty converging to it. (The sequence in Brercise 8 will serve this purpose just fine.) 15. Is the orbit of the point (01 001 0001 00001...) under ¢ dense in 3:7 No—in fact, the orbit of (01 001 0001 00001...) stays away from Mus altogether. And there's really nothing special about the systematically increasing number of Os in this string. No element of {atrings of 0s and Is not having 11 as a substring} hhas an orbit which is dense in 32 16. Is it possible to give an example of an orbit under o that accumulates comes arbitrarily close to but never equals) the two fixed points of o, but which is not dense? Consider the orbit of (01 0011 000111 00001111...)99 CHAPTER 10. CHAOS under ¢. ‘This orbit comes arbitrarily close to either of the fixed points, but itis not dense since it stays away from all other periodic points, for example. 17. Prove that, if's € ¥, there are sequences ¢ arbitrarily close to s for which dfo"(s),0"(¢)) = 2 for all sufficiently large n. Let s = (s00182....) € Zand consider the point (9081 ---Snfngrang2 ---) By the Proximity Theorem, we know that, L 1 < gn but. dfo*(s),o*(t)] =2 for all k >. 18. Prove that the set of endpoints of removed intervals in the Cantor middle-thirds set is a dense subset of the Cantor set Points in the Cantor set have no 1s in their ternary expansion, while end- points of removed intervals correspond to terminating ternary expansions. ‘The rest of the proof is straightforward (see Exercise 2).. 19, Let V(z) = 2[z|—2. Find the fixed points of V and V2. Compute an expression for V°. ‘The function V:[~2,2] ~ [~2,2] is a piecewise linear approximation to our old friend Q-2:[~2,2] —» [2,2]. See Figure 10.3. Note that both V and Q-2 are 2-to-1 and onto the closed interval [-2,2} Now, by definition of absolute value, we have that _fimeca, Weeds vg=["e7, PSs. 9 of this piecewise linear map equal to 2 and solving, we fix V = {-2/3,2}. Exercise 19 100 Figure 10.3: The quadratic map Q_2(2) = 2” —2 and its piecewise linear approximation V(x) = 2{z| Now recall that a formula for V2(x) was derived in Section 10.2 of the text. Its 4e-6 iflges?2 4242 if0SeS1 4e+2 if -1 1 Ve) =V(V%(2)) V(—42 6) Adz -6)—2 =-8r-14 -3/2S2<-1 + O 0,1) commutes. (Note that the doubling map D needs to be defined on the closed unit interval for this to work.) Thus T’ will be chaotic by virtue of Exercise 20. This is because orbits under iteration of D map to dynamically equivalent orbits under T. In fact, we now prove by induction that ToD =™ (10.1) {or all n > 0. Suppose Equation 10.1 is true for n= k, Then ToD = 7% + ToToDI- = ror, and ince ToD = ToT (this will be verified in a moment) we have ToDt= wa which completes the inductive proof. We remark that (10.1) gives an ex: plicit formula for T(z) since we already know that D"-(2) = 2"-12 mod 1 Bxercise 21 104 We now show that D is semi-conjugate to T' via T, or in other words, that ToD=T oT. There are four cases to consider for To T: OSr 0ST(e)<1/2 > TeT(z)=T(22) 222) de WASr S12 > 1/2ST@)<1 4 Tot(z)=T22) 2-222) 2— 4a (2-22) = 2-22-22) 42-2 + 0ST) <1/2 + ToT(2)=T(2-22) 22-22) =4-4e I2srs34 > 12 0 12 W2<2s3/4 > O BASe 1/2 (0,1) | I tin 0 eoentlae viewlliiscndipds Iaxier ssi wlll ten Vor'=Gou. (10.2) Note that U[0,1]-+ [1,1] is @ homeomorphism. Now, the lefthand side of (10.2) is given by _fcoQm) — itoR Bxercise 24 106 commutes. ‘That is, we seck a homeomorphism L such that LeG=Q-20L. Suppose L:Ik +R is of the form L(x) = az +6. Then Lo G(z) = a(2z? 1+ = 2az*—a+b, whereas Q_20L(2) = (az+0)?—2 = a?a?+2abe +022. Equating coefficients, we obtain the system of equations e 2ab Be which has solution a=2 b=0 Let's check this result: LeG(z) = (22? — 1) = 42? —2, and Q-20L(2) = (22)? -2= 427-2. ¥ 24, Prove that Fa(2) = 4z(1 2) is chaotie on (0, 1] Our goal is to find a linear map W:[~2,2] + (0, 1] such that, FroW=WoQ-s Suppose W(2) = az-+6. ‘Then Fyo W(x) = A(ax + )(1— ax 5) = Max — a?s* — abs +6 — aba — 0) = 4((a— 2ab)x— a?a? +6?) (10.3) whereas Wo Qua(e) = a(2?—2) 40 =a2"—2atb, (10.4) Equating coefficients in (10.3) and (10.4), we arrive at the following system of equations: 4a? (10.5) Ao ~ 2ab) = (10.6) 46-8) (102)107 CHAPTER 10. CHAOS Exercise 24 From (10.5), we see that a = 0 or a = —1/4. Plugging the latter into (10.6), we find that 1+ 26 = 0 which says that § = 1/2. Checking these results against (10.7), we have ‘Thus, and the reader may check that F; © W(2) = We Quo(2). Hence, Fy is chaotic by virtue of Exercise 23 where it was shown that Q-2 was chaotic. Indeed, combining the rosults of Exercises 20-24, we have that to —2s fos Hays L299 %s 0, D a le Q-2 Fs toys oy 4s iy 25 2,35 0 ‘commutes. Note that all conjugacies but the first are homeomorphisms. 108Chapter 11 Sarkovskii’s Theorem Exercises 1. Can a continuous function on B have a periodic point of period 48 and not one of period 56? Why ‘Yes, Observe that 56 = 2°+7 precedes 48 = 2.3 in the Sarkovskii ordering. ‘Thus, if @ continuous function F has a cycle of period 56, then it also has a cycle of period 48. Now there is a continuous function with a cyele of period 48 that docs not have a cycle of period 56 (in fact, by the theorem on Pp. 138 of the text, there existe @ continuous funetion with period 48 having. no cycles of any period preceding, 48 in the Sarkovskii ordering) but this is not necessarily s0. In other words, a continuous function F with a periodie point of period 48 may or may not have a periodic point of period 56. 2. Can a continuous function on R have a periodie point of period 176 and not one of period 967 Why? No. In this case, 176 = 24. 11 precedes 96 = 2-3 in the Sarkovskii ordeting, aud so, by Sarkovskii's theorem," a continuous function with a periodic point of period 176 must also have a periodic point of period 96. 3. Give an example of a function F: 0, 1] + [0,1] that has a periodic point of period 3 and no other periods. Can this happen? 7For an inveresting proof of Sarkovski's theorem, see: Harvey Kaplan (1987). A cartoon-asssted proof of Sarkovshis theorem. Aaner. J. Physies 95(11}, 1029-1083, Bxercise 3 110 o $ F a Figure 11.1; A discontinuous piecewise linear map having only period 3 points. ‘Well, Sarkovskii’s theorem applies only to continuous functions and so it follows that such function must be discontinuous. Comsiser the function e+lf3 if0 1, _{2+t/m Sn = (Ee ey has nothing but periodic points of period m. Also show that O<2<(m-1)/m (m—Dfm ot 28 > t 2. 3 (2) Graph of F (0) Graph of Figure 11.2: Two graphs with period 4. T/A 1/4 8/4 9/4 1/4 1/8 + 23/844 7/8 ++ 17/8 1/8 and other 4-cycles are readily found. Indeed, we will show in what follows that G has nothing but 4-cyctes. ‘The equation of the piecewise linear graph in Figure 11.2b is 3-2 if0 (5,6) [2,9] [6,7] Lo [7h But what about [4,5]? Since 14,5] B35] [8,6] + [2,6] 2,7) [1,7] (116) there is indeed a period 5 point in [4,5]. But we claim that there’s exactly ‘one such point in (4,5), and in fact, it’s the fixed point x = 13/3. This is because each of the mappings in (11.6) is strictly decreasing and therefore [4,5] — (1,7 is strictly decreasing (since the composition of an odd number of decreasing functions is decreasing). Thus F has a unique fixed point in [4,5], and moreover, this point must be the fixed point of F Does F have a 3-cyele? No, for ifit did, it would also have a 5-cycle which we've already shown does not exist,bs 3 ; , CEN ‘LEN : (a) (b) Figure 11.4: ‘Two graphs with no odd periods. By the way, there’s a nice pattern here that deserves mention, Period 5, but not period 3: rector terior iest (i) Period 7, but not period 5: Ide Se Seb TAL (118) See how (11.8) is obtained from (11.7)? Just add one to each point in the period 5 orbit and then add the iteration 7++ I at the end. Similarly, we may construct a map with a period 9 orbit, but not period 7. The necessary Scyele would be TSO dre Tie Bre Be DH OE L (19) ‘The map corresponding to (11.9) is not hard to construct, and is left as an exercise, 7. Consider the graph in Figure 14a, Prove that this function has a cycle of period 6 but no cycles of any odd period. Exercise 8 16 his function has the G-cycle Ordre Tree De BH 0 which may be checked easily from the graph, and so F has periodic points of all even periods. ‘The mappings (0,1) 14,5) 11,2) 8,5] [0,2] 6 3,5) RIF DI OH early show the absence of odd periodic points, however. ‘Take period 3 points, for instance. Since F501} = (3,5) (1.10) FUL 13,5) quay 3,5] = (0,2), (11.12) there are no period 8 points in these intervals since [0,2] maps to [3, 5], and vice versa, ‘The only remaining interval is [2,3], but it contains absolutely no periodic points save the lone fixed point of F. So F has no period 3 points. Let's also check period 5 points, Again, there aren’t any since (11.10-11.12) hold with F replaced with F®, In fact, there are no periodic points with any odd period since F410, 1] = [3,5] Fett, 2) = [3,5] F403, 5] = [0,2] for all positive integers n. Here's a totally different approach to this problem: First, verify that the piecewise linear function in Figure 11.5 has the period 3 orbit OIRO. Since this function is continuous, Sarkovskii’s theorem guarantees that it has periodic points of all periods.uz CHAPTER 11. SARKOVSKI'S THEOREM ° 1 Figure 11.5: A continuous funetion with a 3-cycle Now construct: the double of this function? to produce the graph in Fig- ure 11.4a.4 Since the period of every orbit, has doubled, this function has only oven periodie pointe, 8 Consider the fimetion whose graph is displayed in Figure L1.4b. Prove ‘that this function has eycles of all even periods but no odd periods (except 1), Observe, first of all, that 2 is fixed by F.5 Inspection of the graph in Figure 11.4b shows that [3,4] - [0,1] and [2,3] + [1,2], both of which map to [2,4]. But [2,4] [0,2], and viee versa. In general, F401] = FUL, 9] = [2,4] and Festa, 3] = F+p8, 4} 0.2), and hence, there can be no odd periodic points. On the other hand, all odie orbits exist since (0,1) = F™1,2)= 1,2] 08 of RL, Devaney's An Introduction to Chaotic Dynamical Systems Second Baton, Addison-Wesley, 1989, ‘This isnot quite tue, but the discrepancy ix not important Indeed, any multiple ofa power of two in [0,4] is eventually fixed. Exercise 9 18 and F(2,3) = (8,4) = (2,4). ‘Thus we have that [0,1] ¢ (0, 2) 2 (1,2) and [2,3] ¢ [2,4] 2 [3,4] and 20 even periodic orbits are guaranteed to exist 9. Consider the subshift of finite type E’ C E determined by the rules 1 may follow 0 and both 0 and I may follow 1, as discussed in Section 11.4. 9a) Prove that periodic points for o are dense in 2’ Let s = (sosis2...) €B!. Given € > 0, choose an integer n such that 1/2" < ¢, Now consider (ost) and observe that s and ¢ agree up through their respective (n+ I)st centries.® By the Proximity Theorem, dist) 2" 0 be given. "Then there is a third point, namely, (s081 .--Sultotr -.-tn--.) "Note diet t= (oar) will not work since both ao and sy may be sero, and nce gE119 CHAPTER 11. SARKOVSKINS THEOREM ° Figure 11.6: A subshift of finite type on three symbols, which is within ¢ of s, and whose orbit comes within ¢ of t. Note that a 1 has been inserted between the ; and the tj in the event that 0. The tail of the sequence is arbitrary. Incidentally, the previous two facts taken together show that the subshift is sensitive to initial conditions” from which it follows that the subshift is chaotic ‘The following four problems deal with the subshift of P, the space of sequences of 0's, 1's, and 2's, determined by the rules that I may follow 0, 2 may follow 1, and 0, 1, of 2 may fellow 2. Let Sa = { (s08182-..) | 5 € {0,1,2}} and let Ys D Uh ={(s0s180...)|2 must follow 1 must follow 0 } See Figure 11.6. 10. Is this subset of By closed? Yes, Choose s ¢ Bj. ‘Then there exists ak such that both sx and sia are both zero or one. Choose ¢ < 1/2 and let ¢ be any point in 33 such that d{s, t] < ¢. By the Proximity theorem, s and t agree up through the kth position. In particular, sy = te and 94-1 = ty-1, and so te and tha are both zero or one. Hence, ¢ ¢ 34 and so the complement of 3% is open ‘Therefore, 3 itself is closed 11. Are periodie points dense for this subshift? ‘Yes. ‘The proof mirrors the one given in Exercise 9a except here we let t= Gare) TSee J. Banks, J. Brooks, G. Calm G. Davi and P. Stacey (1992). On Devaney's definition of chaos. Anter. Math. Monthly 99(4), 392-24. Exercise 15 120 Figure 11.7: A subshift of finite type with a S-cycle but no fixed or period 2 points, where n is chosen in such a way that 1/2” < ¢ and sy = 2 (which can always be done since s contains an infinite number of 2s). ‘This ensures that € € 4 12. Is there a dense orbit for this subshift? We will show that this subshift is transitive. Let s = (sosis2...) and t= (lolita...) be any two points in Ub, and let ¢ > 0 be given, Choose n such that 1/2" < ¢ and 5, ‘Then any point beginning with the ty ie within € of # and has an orbit which comes eequence soe: tu fots within ¢ of t 13, How many periodic points of periods 2, 3, and 4 satisfy these rules? Period 1: (3) 1 Period 2: (3), (12), (2) 3 Period 3: (2), (12), (120), (201), (122), (221), (212) 7 Pesiod 4: (2), (12), (21), (0122), (1220), (2201), (2012), (2122), (1222), (2221), (2212) " ‘True or false: There are 3+ 7 + 11 = 21 periodic points of period 14. Construct a subshift of finite type in U3 which has a period 3 point but no fixed or period 2 points Consider the subshift: depicted in Figure 11.7. It has but three points, all of which are 3-cycles. 15. Discuss the dynamics of the subshift: given by the directed graph in Figure 11.8a, Are periodic points dense for this subshift? Is there a dense ‘orbit? How many periodic points of period n does this subshift have? ‘There's not much difference between this subshift on three symbols and the full shift on two symbols. The only difference is that strings may start with121 CHAPTER 11. SARKOVSKII'S THEOREM (a) {b) Figure 11.8: More subshifts of finite type. the symbol 2 in the case of the subshift, Indeed, for » € D4, we have that o(s) € Ea, and in fact, per, o is the same as before (the size of Per, 7 has doubled, however). 16. Discuss the dynamics of the subshift given by the directed graph in Figure 11.8b. Are periodic points dense for this subshift? Is there a dense orbit? Except for those strings beginning with 1, there's @ 1-1 correspondence between elements of 3% and 3»: just substitute 0 for every O1, and 1 for each 21, It seems that this subshift is equivalent to the one in Exercise 15. Exercise 16 122Chapter 12 The Role of the Critical Orbit Exercises 1. Compute the Schwarsian derivative for the following functions and de- cide if SF(z) <0 for all. Ja) F(z) = 2%; F'(@) = 22; F"(2) = 2 and F(z) = alee) -ale] SF(2)= oe <0 forall, In fact, for the quadratic family, Q-(z) = 27 +e, 3 SQe(s) =~ 325 for all, Exercise 1 124 1b) F(x) = 23; P(e) = 82% P"(2) = 62; and F"(2) = 6. sr(ay = 2"@)_3 [Fe ro “21F@ ssf) an <0” forall. PM(2) = 96%; and PM(x) = Be le) F(a) = 6; P(e) = _FMe)_8[EMe)y? SE(2)= Bie) ~ 3 [es ate 2 ae ~ | <0. In general, for F(2) =, sr(2)=—-" 1d) F(z) = cos(2? + 1); F'(#) = —2esin(x? + 1); and F"(z) = 42 cos(2? +1) —2sin(z? + 1). Also, PM"(2) = 829 sin(x? + 1) — 8x cos(x? + 1) — 4x cos(x? + 1) B24 sin(x? + 1) — 12x cos(z” + 1). SF(2) ieee Fiz) ~ 2 | P(e) __ 82% sin(2? + 1) ~ 12x co9(2? + 1) ~ =e sina? + 5 ea125 CHAPTER 12. THE ROLE OF THE CRITICAL ORBIT = Genta? +1)~tet— 3 (seaa(ets +2)” = Gcol(2? +1) — da? — 3 (12teosta? 1) + Acot(e? +1) + 3) =~ (42 +627 cot®(x? +0435) <0 forall x We better check our work. Let O(z) = eos and G(2) = 2? +1 80 that C'oG(z) = cos(z? + 1). An easy computation yields and sae 3) he So, by the chain rule for Schwargian derivatives, S(C oG)(x) = SC(G(z)) -[G'()? + SG(z) = (afew +0) tet de = 42? — 62 cot?(2? +1) — a which checks. te) F(z) = arctan; Fi(2) = (142%); and P"(2) = -22(1-4 2%). P(x) = B2"(1+.22)-9 — 2+ 22? 2(1+2%)-8(42? (1+ 2") 2(1 + 2?)-*(32? — 1). sr eg 3 Lee) Bxercise 3 126 = AL 24-8327 -1) ora Sx?-1) 3 __ Was? 4— 122? 2+ 2 “are <0 forall. 2. Ist true that $(F + G)(2) = SF(2) + SG(2)? If 60, prove it. If not, give a counterexample, False. Unlike ordinary differentiation, the Schwarzian derivative does not distribute over addition. Let F(z) = G(2) =e". Then eT) sree seyer(E 3, Ip it true that S(F -G)(2) = SF(x)-C(e) + F(2)-SG(e)? If 90, prove it. If not, give a counterexample. Unfortunately, there is no product-like rule for Schwarsian derivatives. Let F(z) = € and G(2) = &*. By Exercise le, SF(z) = —2 and SG(z) = 9/2. But SF(2)-G(z) + (2) - SG(2) = —2e™ — (9/2)e*127 CHAPTER 12. THE ROLE OF THE CRITICAL ORBIT f -25/2 = S(F -G)(2) since F -G(z) = 4. Is it true that S(cF)(z) = eS (x) where ¢ is a constant? If so, prove it, If not, give a counterexample. Neither multiplicative nor additive constants have any effect on the Schwars- ian derivative. Since (6) = ¢- F for all n, we have that = Mz) _ 3 oka)? SEM) = Tray ~ 3 LePe) _F"@)_3 (Fe) Fe) ~3F@ = SF(2). Also, since (P+) = FC) for all n, SUF + e\(2) = SF(2). 5. Give an example ofa function that has SF(2) > 0 for at lest some Bxercise 4i at the end of Chapter 5 provides such an example. There we found the origin to be a weakly repelling fixed point for F(z) = —2 ~ 23 —2F"(0) — 3[F"(O)}? = 12. (This is the same as computing $(0) = 6, by the way.) In fact, = a3 [tay SPO) = Fey ~ 3 [Fay = 6-362? ya which is positive when 6— 3627 > 0 Exercise 7 128 ‘or when ic This example illustrates but one of the two eases mentioned in the proof of the Schivarzian Min-Max Principle given in the text. 6. Prove that S(1/2) = 0 and S(az +6) = 0. Conclude that $F (x) = where Let R(2) = 1/z and D(z) = az +b. ‘Then R(e) = 1/2, Re) = 2/23, and "(x)= 6/24. Also, L!(2) = a@ and LO(2) = 0 for all n > I. ‘Therefore, and Siz) Applying the chain rule for Schwarsian derivatives, S(io L)(x) = SR(L(x)) -[E'(2)]? + SL(e) +0 Question: Do all totally periodic functions have zero Schwarzian deriva: tive? 7. Compute SM(2) where Mo= 3 We will show that a linear fractional transformation, or Mabius transfor- mation, has zero Schwarsian derivative. Our goal is to find constants &,129 CHAPTER 12. THE ROLE OF THE CRITICAL ORBIT dep, and such that arth yy be ceed ths” ‘Then the Mobius transformation is the composition of functions that we have already shown to have zero Schwarzian derivative (see Exercise 6), Hence, the Mbius transformation itself has zero Schwarsian derivative by repeated applications of the chain rule. Ife # 0, then we may write (12.4) Now, by _ bale ths) +a SATE eth = Mat (hiks + ka) (122) ae Equating coafficiente in (12-1) and (12.2), wo noe immediately that ky = a/e and ky = d/e. Also, Eyky + hy which implies that ‘Thus, ‘and we are done since we have shown that the Mabius transformation is the composition of both linear and inverse transformations. 9. Give a formula for $(F 0 Go H)(2) in terms of SF, SG, SH, and the derivatives of these functions? Exercise 10 130 ‘Two applications of the chain rule for Schwarzian derivatives yields S(F0G)0 2) (F 0 G)(H(2))-[H'(a)]? + S(H\(a) = [SF(Go H(2))-(C'o H(z)? + SG(H(z))] [H(@yP + Si(@) (123) =SF(GoH(2))-((C'o Hz) He)? + SG(H(2)) -[H"(2))? + SH(z) (24) = SF(Go H(a))-((Go H(z) + 5G A(z) (Fo (GoHl)(z). Either of (12.3) or (12.4) solves the exercise, but we have shown more, that i, composition of functions is associative with respect to the Schwarzian derivative, 10, Compute the Scwarsian derivatives foreach ofthe allowing functions: 109) F(2) = sn (072) We would tke to apply the reaulte of Exercie 9. To thie end we begin by letting H(z) = sine. Then H’(2) = cos, H"(2) = —sinz, H(z) =~ cose, and Now let B(2) =e". But Bis its own derivative, and so sea) =5 {| 2 For Q(2) = 2? +2, we have Q(z) = 22, Q"(2! 2, QM(2) =0, and sae) =0-$[2]°1B CHAPTER 12. THE ROLE OF THE ORITICAL ORBIT ‘Ba? Note that He Q(z) = c*"#?, and so in (#2) HoPoQ2)= ‘To apply Exercise 9, we must. compute S(E 0 Q\(z) = Finally, since (Ho Q)'(z) = 2xe*”#?, we see that SH 0 BeQy(z) = SH(E 0Q(e))-(E 0 QY(a))? + 5{E 0 Q\(2) m= (1 eae (ore9)) [sate] (ae? 8.) <0 forall 10d) Fle) = seabray at Ole) = ene. Br tharos 8, s0(@)= 1 Beare Now let Q(2) = 222. By Exercise 1a and the fact that S(F-+c) = SF (which was shown in Bxereise 4), we see that 500) =-35,. In fact, SQ.(z) = —3/(22*) for all ¢. Now consider CoQ. By the bain oe Sr Scnepnsian dsivaee, S(C 2 Q(z) = SC(Q(#)) -1Q"(@)P + SQ(z) 3 = fod Seaitis* alhat_ 2 = (1 Fore’ »)4 a 2 (12 r420e-9 434) Exercise 10 132 which is practically the same answer obtained in Exercise 1d. At any rate, the last step is to let R(2) = 1/2. By Exercise 6, R(C Q(z) -[(CoQ) (2)! + S(C 2 Q)z) (Co Q)e) S(Ro Co Q\e since SRExercise 1 134 Chapter 13 Newton’s Method Exercises Figure 19:1: The Newton LF. for F(a) = 42x is constant 1. Use graphical analysis ta completely describe all anbits ofthe associated Newton iteration function for F. 1a) F(e) 2 PM) NF( = (4—22)/(-2) =e - (2-2) =2. All orbits are eventually fixed on the fixed point 2 after one iteration, See Figure 13.1 Ib) F(z) +2222 > P(x) = 22-2 > PM2) =2, 4-22 => P(e) Figure 13.2: The Newton LP. for F(2! 2c is a typical example. All orbits, except that of 1, converge on a root. ‘The bi ent from the graph (see Figure 13.2). ns are appar-135 CHAPTER 13. NEWTON'S METHOD Figure 13.3: The Newton LF. for F(z) = 2/9 is linear. Le) F(a) = 2 P(e) = Ba? > P(e) = a9 NF(2) = 2 is =2-82/2 =-2/2 Observe in Figure 13.3 that F is not differentiable at the origin, yet [NF has an attracting fixed point there. In fact, all orbits are attracted to 0, but not quadratically ‘This example is typical of an entire class of functions, namely, Fa(2) = 24 with 1/2 I, the derivative exists (but vanishes) and the Newton iteration function still converges. 1a) F(2) = 24 +22 = 2%(2? +1) Fi(e) = 409 + 2a = 2e(22" + 1), PMs) = 120? +2. atta? NP@)=8~ Tse te Exercise 1 136 Figure 13.4: The Newton LF. for F(z) = 24+ 22 det 422" — (ot 4 22) ait _ Sate “Bee Curiously, the graph of NF depicted in Figure 13.4 is almost linear. Indeed, Bette et! de aa aye TTT aud snc " F@)Fz) OOO RET (2? + 1)(122? +2) 427227 + 1 (2? +1)(62? +1) “Watt? we have that (NFY(0) = (1)(1)/2 = 1/2. Recall, however, that if p is a root of multiplicity m, then (NFY'(p) = (m—1)/m, and since m = 2 in this ease, the previous calculation is verified, le) Fe) = Wz 2 => M2) = 22%, But F has 1 > F(a) =137 CHAPTER 13. NEWTON'S METHOD (@) The Newton LF. for F(x) = 1/s is. (b) The Newton LP. for F(z) = 1/s~ 1 Tinear. isa parabola, Figure 13.5: Two members of the family F(z) = I/z-+e. no roots, and NF(2) ate = 28. Hence, all orbits, except the orbit of O, diverge. (See Figure 13.5a.) If) F(z) = 1/21 (1~2)/2. (Note that (2) vanishes when x = 1.) 2 setae’ = 2x2? =2(2-2) In Figure 13.5, geaphical analysis suggests that {2|(NF)%(2) > Las n +00) ={2]0<2<2)} Exercise 1 138 Ig) F(a) = 2/vT4 a = 2(1 4 22)? By the product rule, PY(2) = (14-22)? + a(—3)(1 + 22)-*72(22) 1+ 22)-9/21(1 + 28) — 23] = (+2 ‘The Newton iteration function for F is a(t aty-tl2 NF(2) =2- ee ~a(1 +2") and s0 DNF(2) = —822. By inspection of Figure 13.6, we see that (NF}*(1) = 1, and 90 1 is a period 2 point for NF. But the corre- sponding 2-cycle is repelling since D(NF)*(1) = DNF(1)- DNF(-1) = 3? -(-3)(-? =9. From the graph, it appears that the open interval (1,1) is the im- ‘mediate basin of 0. Indeed, observe that NF (2) = —2° (NF (2) = —(-28)8 (NF)(2) = ~(-2°Y (wPy"(2) = (12 and [(VF)*(z)| = [(—1)"23"| = [2*"| +0 as n + 00 provided [a| < 1. Ih) F(a) = act > Pile) = ef +2ef = ef + F(a). Consequently, erat tet patel — 20 epee “Stee “The NF(z) = 2139 CHAPTER 13. NEWTON'S METHOD Exercise 3 140 ‘The Newton LF. for F(z) = 2/VT+27 is a simple cubic Figure 13, polynomial ZN Figure 13.7: The Newton LP. for F(z) = 2e* ‘The reader should verify the following observations taken from Fig- uure 18,7, oome of which ae trivial: 1. F has a minimum at @ = 1 2,2 —1 is an asymptote for NF. 3. F(z) +0" as = ~00. 4. NF(2) + 2 95 2 + -00. 5. NF(z) > £ as 2 + +00. 6. 7 For ~00 < 29 <=1, (NF)"(20) = —ce as n 00, For —1 < #9 < 00, (NF)"(20) 0 as n + 00. 2. What happens when Newton's method is applied to R(z) = 3? For R(x) = Vi, we have R'(2) = 1/(2V2). Thus, vi NR(x) ~ Teva) -2e Figure 13.8: The Newton LP. for R(x) = YF is linear. Sce Figure 13.8. Now we know that per, Nit = BR since (WR)*(z) = = for all 2, and s0 Newton’s method fails miserably in this case. Why? Mest likely because 2'(0) does not «ui CHAPTER 13, NEWTON'S METHOD 3. Find all fixed points for the associated Newton iteration function for P(e) = £/(2= 1)" when m = 1,2,3,... Which are attracting and which are repelling? We write F(z) = 2(2— 1)" and comput Pia) = (@— 17 — a(x — 1)" = (2-1) "(2-1 m2) 2(1— m) — (w= 1) 0 or when 2 = 1/(1— m), provided the product rule which vanishes when 2(1—m)— m #1. Likewise, Pa) = (= m)(x = 1-9 — Om 41)(2(1 =m) = le = 1) (e = 1-49 — mw = 1) — (m+ 1)(@(L =m) ~ 1) = (@—1)-47)(a(mm? — ma) 42m) which well use later in this exercise. Now the Newton iteration function for F is _ #e) F(z) NF(2) = 2 ae- 2-1)" 2° =m) = He F _ 2(2(1 =m) ~1)- 2-1) z(1—m)—1 (a3. Setting Equation (13.1) equal to » and solving, we find that fix NF = {0,1}. Note carefully that these fixed points are independent of m. (See also Figures 13.9 and 13.10.) But are they attracting, superattracting, repelling, or what? ‘To answer this question, we compute FF") Fey DNF(x) = Exercise 3 142 Figure 13.9: The Newton LP. for F(z) = 2/(e—1) is the parabola z + x? (m? — m) + Ime =e =m) 1 (a3) ‘Thus, DNF(0) = 0 and 2 =m) FI oe. T= my = m= 24) amt “m1 which is greater than unity for all m > 1. So the latter fixed point is always repelling. Now it appears as though NF has two critical points when m = 3 but ‘only a single such point for m = 2 (see Figure 13.10). Let's examine this conjecture more elosely. Setting the derivative in (13.2) equal to 0, we have ma(x(m — 1) +2) Gr 1 > ma(e(m—1)+2)=0 4 220 of 2=2/(1-m), and so indeed, there are two critical points for all m > 1. But why isn’t143 CHAPTER 13. NEWTON'S METHOD (yma? bm Figure 13.10: Two more members of the family F(x) = x/(z— 1)". the second one visible in Figure 13.10a? Well, for # = 2/(1—m), we have 4m NF@)=-Gs and so for m = 2, the critical point is at (~2,—8). But this point is beyond, ‘the scope of the graph in Figure 13.10a, and sure enough, zooming out we set the more complete picture seen in Figure 13.11. 4, Consider the Newton iteration function for F(x) = seca. What are the fixed points for NF? Does this contradict the Newton Fixed Point ‘Theorem? Why or why not? We begin with the following calculations: P(e) =secx; ——PM(2) = see sec? 2 + sec z tan ztanz P(g) = secs tana; = soe 3(s00? x + tan? 2). Hence, F(z) NF(e) = =~ Frey Exercise 5 144 Figure 13.11: A distant view of F(z) = 2/(2 = 1)? secx fecwtane cot = hhas fixed points at the zeros of the cotangent function, that is, at x for wonzero integers k. This result is unexpected however, since I" itself has no zeros whatsoever, But observe that these are precisely those values for which secant is undefined. Moreover, F(R") ) sec? (sec? x + tan? 2) (Gece tan 2) a ltete (Fy) which may also be obtained by direct differentiation of (13.3). Now, cose- cant obtains relative optima at the fixed points of NF (or the 2er08 of I). In fact, ese(kn/2) = 1 for nonzero integers k. So, (NFY(kx/2) = 2> 1, and s0 these fixed points are repelling under iteration of NF. 5. Suppose P and @ are polynomials and let F(2) = P(2)/Q(2). What can be said about the fixed points of the associated Newton iteration func- tion for F? Which fixed points are attracting and which are repelling?145 CHAPTER 13. NEWTON'S METHOD P(x)/Q(x), then F@)= (2) P(2) - O(2)P(2) @@y and so FO) Fe) P()/Qe) OEP R)— 1@PDMO? P(}Q(2) WP) = VOPR) “The fixed points for NF are found by solving 2-—— P@@a) WGP) -I@PE) ‘which implies that P(2)Q(2) = 0. In other words, we have found that fix NF = {| P(2)Q(#) = 0). ‘Throughout the rest of this exercise, define A, (2) = Q(x) P"(#)—Q"(z)P(z) and Ba(2) = Q(2)P"(2)-Q" (2) P(x) for convenience. Now it can be shown tat Q(#)Aa(2) — 20" An) (Q(z)* PMe)= and 80 eV) — Fee) (eye) = Trey = BB Meeps wir = Plt)Q(z)Aa(z) — 2P(2)Q"()Ar(z) Ai(2) But when P(z)Q(2) = 0, Equation (13.4) reduces to = 2PC@)Q'e)Aul2) afte) (13.4) (NFY(z) Exercise 6 46 ‘We may summarize all of these results as follows. For an arbitrary r funetion F(z) = P(z)/Q(z), we have that fix WF = {| P(z)Q(2) = 0), each of which is attracting if and only if, | =2P(z)Q'(z) Q(z)P(z) — Q(z) P(2) Inverting both sides of (13.5) and simplifying, we have the equivalent con- dition | Qe) PG P(2)Q"(2) 6. A bifurcation. Consider the family of functions F,(2) = 2?+y. Clearly, F(z) = 0 has two roots when <0, one root when 1 = 0, and no real roots when > 0. Your goal in these exercises is to investigate how the dynamics of the associated Newton iteration function changes as changes. <1 (13.5) >2 a) Sketch the graphs of the associated Newton iteration function NF, in the three cases <0, = 0, and 4 > 0. ‘We handle the simplest case first. When = 0, 2 zoe WR(2)=2- 5 =2-F= 5. See Figure 13.12 for the ease = 0. In general, eee oe NEG z ‘Typical members of this family of I.F.s are shown in Figure 13.13. 6b) Use graphical analysis to explain the dynamics of NF, when <0 and = 0. ‘The dynamics of NFp are simple: the origin attracts all orbits, but not quadratically. On the other hand, the Newton LP. for Fy for x <0 is more interesting. ‘The orbits of all positive numbers quadratically converge to the positive fixed point, whereas the orbits of negative numbers converge to the negative fixed point. See the graphical anal- yes depicted in Figure 13.147 CHAPTER 13. NEWTON'S METHOD Figure 13.12: The Newton iteration function for F(x) = 2" is linear. fyn 0, the Newton iteration function for x + 2? + Lis conjugate (o the Newton iteration function for Fj, vie the conjugacy H(z) = fz. Conclude that the Newton iteration function is chaotic for u> 0. ‘As suggested, we let > 0 and show that NA te w | commutes. This follows sinee nonne =a (224) NF, 0 A(x) =149 CHAPTER 13. NEWTON'S METHOD Exercise 7 150 an vere) =n(2? +1) eres 241 Re 7. A more complicated bifurcation. Consider the family of functions given by Gy(z) = 272-1) +h. Ta) Sketch the graphs of the Newton iteration function NG), in the three cases <0, t= 0, and jt > 0. Figure 13.15: Chaotic dynamics of NF, for 1 > 0. We have that Gi,(z) = 2(8z — 2) and Gi(2) = 6x ~2, and in fact, all of Gy’s derivatives are independent of 4. Now, _ Wary 3 ea Dae NGie) a2) =e? = 1) 2(32— 2) - 2%(e—1)— Dye 223) _ peel 2222-1) = =i 2(5E= 9) 6d) Find an analogous conjugacy when jt <0 We will show that In particular, when je NGo(e) = 22D is shown in Figure 13.16. Other representatives of this fi of LPs are shown in Figure 13.17. Tet’s also compute ‘commutes, where H(z) = 2. Again, this is so because (nGyy (a) = SACRE) wonrs(e)=1 (St) = 241 Ge)? — (z= 1) + n\(62 = 2) Be We = ee = 2 (156) and which we'll use below. NF, 0 H(2) = NF,(/=fi2) Tb) Use graphical analysis to discuss the fate of all orbits in ease je151 CHAPTER 13. NEWTON'S METHOD Exercise 7 152 Figure 13.18: Dynamics of NGo(2) = (2(2# —1))/(8x — 2). Figure 13.16: The Newton iteration function for Ga(z) = 2%(x ~ 1) [t's clear from Figure 13.16 that fix NGy = {0,1}. (These fixed points ‘are most easily obtained as the zeros of Go.) Now the derivative of ‘NGo may be computed dizectly from (13.6): Fo) (a) = = Wi62— 2) NON 0) = ae 3 ‘Thus, 2 = 1 is superattracting since (NGo){(1) = 0, but 2 = 0 is not since (NGa)"(0) = 1/2. This is because O is root of multiplicity two of Go(z) = 0. It can be seen from Figure 13.18 that the orbits ofall points to the right ofthe vertical asymptote converge (quadratically) to = 1, while those to the left of the asymptote converge (linearly) to x =O. Since 2 = 2/3 has no preimage, these statements ate true without exception. Te) Show that NG, has exaetly one critical point which is not fixed for all but one p-value. ‘The critical points of NG, are precisely the zeros of Gy, as well as the zeros ofits second derivative. The latter gives rise to the ertical point (nro Wuof (i Basa) =o 11-5844 114? =0 o which, by the quadratic equation, has solutions B84 VST TT 4, See _ 29 VIO TH Of these two roots, only A— produces a real value when plugged into (13.9). See Figure 13.21 for the corresponding Newton iteration functions, Notice Exercise 11 158 () A= (94. V/ () A= @9- VR Figure 13.21: A pair of Newton LPs for F(2) = (2? —1)(2 + A). the striking similarity between Figure 13.21b and Figure 13.19b— there's a conjugacy lurking about! 11. Prove that the equation F(z) = 0 has a root of multiplicity m at. 29 if and only if F(z) may be written in the form F(z) = (@— 20)"G(x) where G does not vanish at x9. Hint: Use the Taylor expansion of F about ‘The hint will be useful going in one direction only. By definition, DF (z0) vanishes for j = 0,1,...,m—1, whereas D™F(20) does not. Now by ‘Taylor's theorem, there exists a € (which depends on z) between 2 and zo such that F@)= Pee aor"! Ge _ myn (DF Cea) , DH PE) = (enn (Pee, PFO 20) = (2 20)"G(2)159 CHAPTER 13. NEWTON'S METHOD where G(20) =D" F()/m! does not vanish. Going the other way, suppose F(z) = (2 20)™G(2) with Gl2o) £0. By Leibniz’s rule, D*F(2) (oe = 20)" D"-4G(a). But for n= 0,1,...,m—1L, we have that D* F(z0) = 0 since vanishes at x = x9 for j = 0,1,...,m—1. On the other hand, D™(z — 0)" = []%p'(m — i) = mt, and so D" F(z0) = m!G(20) which doesn’t vanish by hypothesis, 12. Let G(e) = exp(1/z?) if x # 0 and set G(0) = 0. Compute the Newton iteration function for G. What can be said about the fixed point of NG? Why does this occur? Straightforward calculation yields ee) = BBG _ 2048 and ore) = BOL= Ate _ Gla) = 62*G(a) = 1a@)=8 = ote) $= (13.10) ‘Thus, NG(e) = 2— me = Note that x = 0 is the only fixed point of NG, and Bxercise 12 160 Figure 13.22: The Newton LP. of G(2) = exp(—1/2%) has a weakly attract- ing fixed point at the origin by direct calculation. But (NG)'(0) = 1 and so we must apply the results of Chapter 5. ‘To this end, note that (NG)"(0) = 0 and (NGY"(0) = -3 <0. Hence, by Theorem 5.2, the origin is weakly attracting. But why does this happen? Tt should be clear from the graph in Figure 13.22 that G'(0) = 0. Indeed, the graph appears very flat in the vicinity of the origin. In fact, D"G(0) = for all n, an amazing fact which we now show analytically. By definition, (0) = jn GOH = GO) = Jim SPV) oii 1 ab Trexp(l7i) 1 = lim eye (Nhe! =k t = ey