Andrew Brouwer MA 563: Dynamical Systems Homework 1 Note: I will use B(x, ) and B0 (x, ) to denote the neighborhood

of x and the punctured neighborhood of x, respectively. TI.2 Let f be a C 1 map on R. Let p be a xed point of f . Prove that if |f (p)| > 1, then p is a source. Proof. Assume the hypothesis. As |f (p)| > 1, there is a R such that 1 < a < |f (p)|. By denition, |f (x) f (p)| |f (p)| = lim xp |x p| By denition of limit, there is > 0 such that if x B0 (p, ) then in particular
|f (x)f (p)| |xp| |f (x)f (p)| |xp|

(a, 2|f (p)| a), and,

> a.

)f Let > 0 such that < . Let x0 B0 (p, ). Then |f (x00 p|(p)| > a, that is, |f (x0 )f (p)| > a|x0 |x p|. Let N N such that aN |x0 p| > , which is valid as a > 1. Suppose xn = f n (x0 ) B(p, ) for each n < N (otherwise, were done). Thus, for each n < N , xn B(p, ), |f (xn ) f (p)| > a|xn p|. Thus, as f (p) = p and xn = f (xn1 ) , |f n (x) f (p)| > an |x0 p| for n < N by nite induction. Consider |f N (x)p| = |xN p| = |f (xN 1 )f (p)| > a|xN 1 p| > aaN 1 |x0 p| = aN |x0 p| > . As |f N (x) p| > , f N (x) B(p, ). Thus, by denition, p is unstable.

E1.1 Let l(x) = ax + b where a and b are constants. For which values of a and b does l have an attracting xed point? A repelling xed point? If p is a xed point of l, p = ap + b. If a = 1, then l has a xed point only when b = 0, and, in that case, every point of the domain is a xed point because l is the identity function. Because |l (p)| = 1 for each of these points, we are unable to determine stability by the usual method. However, a further investigation reveals that these xed points fulll the denition of neither attracting nor repelling points. These points are Lyapunov stable, but not asymptotically stable.
ab b If a = 1, then l has a xed point at p = 1a . We see that |l (p)| = 1a . Performing algebraic manipulations on this expression and using Theorem 1.5, we develop the following table:

a b 1 1 a>0 1 a <b< a 1 1 1 a < 0 1 a > b or a 1 < b 1 1 a > 0 1 a > b or a 1 < b 1 1 a<0 1 a <b< a 1 a=0 bR

p attracting attracting repelling repelling attracting

Our algebraic tools break down when |l (p)| = 1, and so we must use some numerical tools to esh out the rest of the chart. Numerical calculations suggest the following: a 0<a<1 b=1 a>1 b=1
1 a 1 a

b or b = or b = 1

1 a 1 a

1 1

p attracting repelling

E1.2 a) Let f (x) = x x2 . Show that 0 is a xed point of f , and describe the dynamical behavior of points near 0. b) Let g(x) = tan x, /2 < x < /2. Show that 0 is a xed point of f , and describe the dynamical behavior of points near 0. c) Give an example of a function h for which h (0) = 1 and x = 0 is an attracting xed point. d) Given an example of a function h for which h (0) = 1 and x = 0 is a repelling xed point. a) We see that 0 = 0 02 , and so 0 is a xed point. Negative points are repelled from 0 whereas positive points are attracted to it. b) We see that 0 = tan(0). Although | sec2 (0)| = 1, 0 is a repelling point. c) Consider f (x) = x3 x, for which f (0) = 0. We see that |f (0)| = |3(0)2 1| = 1, but 0 is an attracting point. d) We have already seen that tan(x) ts this description. Consider f (x) = x3 + x, for which f (0) = 0. Although |f (0)| = |3(0)2 + 1| = 1, 0 is in fact a repelling point. C1.2 Redo Table 1.2 with the logistic map g(x) = 3.5x(1 x). What periodic behavior occurs. Try dierent values of a < 3.57, and report your results. See attached for an Excel spreadsheet of a table with a = 3.5. There appears to be a period-4 orbit of (0.827, 0.500, 0.875, 0.383) (to three decimal places). Further, a=3.51-3.54 have period-4 attractors whereas a = 3.55 3.56 have period-8 attractors. E1.4 Let x1 , . . . , x8 be the eight xed points of G3 where G(x) = 4x(1 x). Clearly, x1 = 0. a) For which i is xi = 3/4? b) Group the remaining points to two orbits of three points each. a) From the graph 1.10(c), we see that x6 = 3/4. b) We examine the itineraries of each of the period three points: x2 x3 x4 x5 x7 x8 LLR LRR LRL RRL RLR RLL

It must be that G2 (x2 ) = G(x4 ) = x8 and G2 (x3 ) = G(x5 ) = x7 . E 1.5 Is the period-2 orbit of f (x) = 2x2 5x on R a source, a sink, or neither? We solve x = f 2 (x) = f (2x2 5x) = 8x4 40x3 + 40x2 25x. This equation simplies to x 5x3 + 5x2 + 3x = 0. Because 0 and 3 are xed points, x and x 3 will be roots of this polynomial. Using polynomial long division to deate this polynomial, we are left with 2 2x 1 = 0. x The quadratic equation is used to determine that the other two roots are 1 2. Since 0 and 3 are the xed points of f , the other two form the period 2 orbit of . We next test the stabilf ity the orbit: |f (p1 )f (p2 )| = ||f (1 + 2)f (1 2)| = |(4(1 + 2) 5)(4(1 2) 5)| = of |(4 2 1)(4 2 1)| = | 31| > 1. Thus, this period-2 orbit is a source.
4

TI.6 Consider the family of maps ga (x) = ax(1 x). Show that if 1 < a < 3 then x = a1 is a sink a and that if a > 3, then x = a1 is a source. Show that if 3 < a < 1 + (6) then ga (x) has a stable a period two orbit, and that if 1 + (6) < a, then the period two orbit is unstable. First, we consider the xed point of ga (x). We solve p = ap(1 p) to discover two xed points 0 and a1 . We see |f (0)| = |a 2a(0)| = |a| > 1 when a > 1, making 0 a source for those functions. a Next, we consider |f ( a1 )| = |a 2a( a1 )| = |2 a|. If 1 < a < 3, then this point is stable. If a a a > 3, it is unstable.
2 Now we consider ga (x) = a2 x (a2 + a3 )x2 + 2a3 x3 a3 x4 . If we set g 2 (x) = x to solve for the period-2 points, we may solve the equation (a2 1)x (a2 + a3 )x2 + 2a3 x3 a3 x4 = 0. Knowing that 0 and a1 are roots, we may deate the polynomial by dividing by x and x a1 . We are left a a with a2 x2 a2 x ax + a + 1 = 0. The roots of this polynomial, assuming a = 0, are a+1 1 2 a2 2a 3 2 a

These roots are real when a > 3. To determine the stability of the period two orbit formed by these points, we perform the following calculation: (x1 )g (x2 )| = |(a(1 2x1 )(a(1 2x2 ))| = |g a+1 a+1 1 + 1 a2 2a 3 2 a2 2a 3 2 2 2 2 2 |a (1 2x1 2x2 + 4x1 x2 )| = |a (1 2 2 + a a a+1 a+1 1 1 + 2 a2 2a 3 2 a2 2a 3 2 = |a2 (a2 + a + a a2 2a 3) (a2 + a 4 2 a a a a2 2a 3) + 4a + 4)| = | a2 + 2a + 4|. We can then solve for when | a2 + 2a + 4| > 1 and when |a2 +2a+4| < 1. Solving the former algebraically, the orbit is unstable when a2 +2a+3 > 0 or when a2 + 2a + 5 < 0. The rst of these two new inequalities tells us that the orbit is unstable between -1 and 3. However, as the orbit does not exist when a < 3, information is unhelpful. this The other inequality tells us that the orbit is unstable when a < 1 6, which again is unhelpful, or when a > 1 + 6. The inequality | a2 + 2a + 4| < 1 conrms that the orbit is stable for 3 < a < 1 + 6. C1.3 What qualitative or quantitative conclusions can be made about the speed of convergence to the period two orbit as a gets closer to a = 1 + 6 3.45? What happens to iterations beginnging at a period two point for a slightly larger than a? See attached for the output of the Excel worksheet. Our numerical investigation suggests that the speed of convergence to the period two points decreases as a a . For a slightly larger than a , we see that a period four orbit has taken the place of the period two orbit. E1.12 The map g(x) = 2x(1 x) has negative values for large x. Population biologists sometimes prefer maps that are positive for positive x. a) Find out for what value of a the map h(x) = axex has a superstable xed point x0 , which means that h(x0 ) = x0 and h (x0 ) = 0. b) Investigate the orbit starting at x0 = 0.1 for this value of a using a calculator. How does the behavior of the orbit dier if a is increased by 50%? c) What is the range a 1 for which h(x) has a positive sink? a) First, we examine the seoncd condition for superstability, namely 0 = h (x0 ) = aex0 ax0 ex0 . Because ex = 0 for any x, this simplies to a = ax0 . Either a = 0 or x0 = 1. If a = 0, 3

then x0 = h(x0 ) = ax0 ex0 = 0. Further, h (x0 ) = aex0 ax0 ex0 = 0(ex0 xex0 ) = 0, and so x0 is superstable. If, on the other hand, x0 = 1, we solve the xed point condition 1 = ae1 . In this case, a = e. To summarize, h(x) = axex has a superstable xed point at 1 if a = e. b) The orbit of 0.1 under the function h(x) = xe1x goes to 1. If we increase a by 50%, that is, if we use the function h(x) = 3 xe1x , the orbit of 0.1 converges to 1.405465. . . . 2 c) Let a 1. If x is a xed point, x = h(x) = axex . We see that 0 is a xed point. We compute |h (0)| = |ae0 a(0)e0 | = a 1, and so this xed point is not a sink. If x = 0, we solve 1 1 1 = aex to nd x = ln a . Now, since we wish this sink to be positive, we set ln a = x > 0, and 1 we nd that x > 0 when a > 1. Thus, h(x) has a positive sink at ln a when a > 1. T1.7 Find the period two orbit of 4x(1 x). Let G(x) = 4x(1 x). Then G2 (x) = 16x 80x2 + 128x3 64x4 . To nd the period two points, we set x = G2 (x), and deate by the factors which represent the xed points of G(x), viz. x 0 and 3 . This results in the polynomial 0 = 16x2 20x + 5. The roots of this polynomial are x 4 5 85 . These roots form the period two orbit of G(x). 8 E1.8 Let G(x) = 4x(1 x). Prove that there are points in [0, 1] that are not xed points, perodic points, or eventually periodic points of f .
ai 1 Consider 3 . Each point in the orbit of 1 will be rational and can be represented as xi = b1 . We 3 n+1 and an+1 = 4an (bn an ). Now, the term (bn an ) can only be divisible see that bn+1 = b2 = 32 n by three if an is divisible by three. Now, an is divisible by three only if 4 an1 (bn1 an1 ) is divisible by three. This argument continues. Each ai and (bi1 ai1 ) is divisible by three only if ai1 is divisible by three. However, as a0 = 1, no ai is divisible by three. Hence, each xi = aii = 3a2ii b is not reducible. Thus, as each xi , in simplest form, has a denominator dierent from each point previous to it in the orbit, there can be no periodicity in this orbit. Hence 1 is a point that is not 3 xed, periodic, or eventually periodic.

T1.10 Let f (x) = 3x mod 1. Prove that a point x is eventually periodic if and only if it is rational. () Suppose x is eventually periodic with period p. Then there is N N such that if n N , then f n+p (x) = f n (x), that is, 3n+p x =mod 1 3n x. That is 1 divides 3n+p x 3n x = 3n (3p 1)x. So, z there is an integer z such that z = 3n (3p 1)x. Thus, x = 3n (3p 1)) , and since z Z and n, p N, x is rational. () Suppose that x0 Q [0, 1). Then x = a for some a Z and b N. As x [0, 1), n b 0 a < b. Let xn = f n (x0 ) = 3n (x0 ) mod 1 = 3 b a mod 1. Then xn may be considered to have the form xn = abn where 0 an < b. Now, observe that ab must be equal to some ai where 0 i < b because there are only b possible values that ai may take. Let N be the least value such that aN is equal to some ai , 0 i < N . Then p = N i is the period of the orbit {xi , xi+1 , . . . , xN 1 }. Thus, for any n N , f n+p (x0 ) = f n (x0 ), and x0 is eventually periodic.

T1.11 Construct the periodic table for f = 3x mod 1. Period N. xed points N. xed points with lower period 1 2 0 2 8 2 3 26 2 4 80 8 5 242 2 6 728 32 N. orbits of this period 2 3 8 18 48 116

T1.12 Let f (x) = 3x mod 1 on [0, 1). Dene distance by the circle metric, i.e. d(x, y) = min {|x y|, 1 |x y|}. 1 a) Show that the distance between any pair of points that lie within 6 of one another is tripled. b) Find a pair of points whose distance is not tripled by the map. c) Show that to prove sensitive dependence for any point, d can be taken to be any positive number less than 1 and that k can be choosen to be the smallest integer greater than ln(d/|xx0 |)/ ln 3. 2
1 a) Let x, y [0, 1) such that = d(x, y) 6 . Take the homeomorphic image of the circle that identies y with 0 and x with . Because 1 , 3 1 , and so 3(mod 1) = 3. As homeomorphism 6 2 preserves distance, |f (x) f (y)| = |f () f (0)| = 3 = 3d(x, y).

b) Consider

1 4

1 and 1 . We see that d( 4 , 1 ) = 2 2

1 4

3 1 and that d(f ( 1 )f ( 1 )) = d( 4 , 2 ) = 1 . 4 2 4

1 c) Let x0 [0, 1). Let 0 < d < 2 . Let > 0. Take N N such that d < 3N . Take d x = (x0 + 3N )(mod 1) . Then x B(x0 , ). Let k be the smallest integer greater than ln(d/|x x0 |)/ ln 3 = ln(3N )/ ln 3 = N . That is, k = N + 1. Then |f k (x) f k (x0 )| = (3k x 3k x0 )(mod 1) = 1 (3k (x0 + 3d ) 3k x0 )(mod 1) = 3d(mod 1) = 3d as d < 2 , and so |f k (x) f k (x0 )| > d. N

T1.13 Prove that for any map f , a source has sensitive dependence on initial conditions. Let x0 be a source of map f . Let > 0. Let d = /2. Let x B0 (x0 , ). By denition of source, there is N N such that f N (x) B0 (x, ). That is, |f N (x) f N (x0 )| = |f N (x) x0 | > > d. Thus, x0 has senstive dependence on initial conditions. T1.14 Find a point that lies in each subinterval of length three of G(x) = 4x(1 x). Itinerary Example LLL 0.01 LLR 0.10 LRR 0.20 LRL 0.40 RRL 0.60 RRR 0.80 RLR 0.90 RLL 0.99

E1.6 Dene the map f (x) = 2x mod 1 on the unit interval [0, 1]. Let L denote the subinterval [0, 1/2] and R the subinterval [1/2, 1]. a) Draw a chart of the itineraries of f . b) Draw the transition graph for f . c) Establish sensitive dependence for orbits under this map. Show that each point has neighbors arbitrarily near that eventually map at least 1/2 unit apart. a) 0L1/2R1 0LL1/4LR1/2RL3/4RR1 0LLL1/8LLR1/4LRL3/8LRR1/2RLL5/8RLR3/4RRL7/8RRR1 b) Same as for the logistic map; all itineraries are possible. c) Sensitive dependence is easy to show; the second part is less obvious. For a given k, the subintervals of sequence k each have length 21k . Let x0 [0, 1], and take d = 1 . Let > 0. Find 4 N N such that 2N > 1 . Identify which subinterval of sequence length N + 2 to which x0 belongs. Identify the subinterval of length N + 2 that lies at least 1 from that of x0 (one can pair LL/RL 4 and LR/RR for example). Because each subinterval of sequence length k has length 21k , a point in the desired subinterval is at most 2N1+1 < 21 < from x0 . Any point in that subinterval, after N N iterations, will be at least 1 from f N (x0 ). Thus x0 has sensitive dependence. 4 To show that we may use d = 1/2, the previous argument suces for most points. It remains to show that any point x0 whose itinerary is LRRR . . . up to length m must eventually have another L at some term k > m. (Similiarity will give us the case for a point whose itinerary is RLLL . . . up to m). This will show that there are 2k1 subintervals between that of x0 and a point who itinerary k1 1 is k Rs, and thus are at least 2 2k = 2 apart. To prove our point, let > 0 and suppose x0 is within of 1/2 on the left side. Let = 2 2 0 . Take N N such that 21 < Thus, there must be N a subinterval of sequence N to the left of the subinterval that contains x0 . With the similarity of argument to the right of 1 , we are nished. 2 E1.7 Dene the tent map on the unit interval by T (x) = 2x if 0 x 1/2 2(1 x) if 1/2 x 1
1

Divide the unit interval into two appropriate subintervals and repeat parts a c of E1.6 for this map. Complete a periodic table for f for periods up to 10. a) Itineraries: Let L denote the subinterval [0, 1/2] and R the subinterval [1/2, 1]. 0L1/2R1 0LL1/4LR1/2RR3/4RL1 0LLL1/8LLR1/4LRR3/8LRL1/2RRL5/8RRR3/4RLR7/8RLL1 b) Again, all transitions are possible. c) The argument is the same as for the previous problem, though the names of the itineraries will be dierent.

d) Periodic table Period N. xed points N. xed points with lower period 1 2 0 2 4 2 3 8 2 4 16 4 5 32 2 6 64 10 7 128 2 8 256 16 9 512 8 10 1024 34 N. orbits of this period 2 1 2 3 6 9 18 30 56 99

Because each period has the same number of xed points as the logistic map (2n ), this periodic table is exactly the same as that of the logistic map. T1.15 Continue the schematic diagram of itinteraries for G(x) = 4x(1x) for four sequence lengths. Itineraries switch at the critical points of G5 (x), i.e. at points where the derivative is zero. The itineraries, in order, are as follows: LLLLLLLRLLRRLLRLLRRLLRRRLRLRLRLL RRLLRRLRRRRRRRRLRLRLRLRRRLLRRLLL T1.16 Let G(x) = 4x(1 x). Let x0 be a point in the subinterval RLLRRRLRLR. a) Is x0 less than or greater than 1/2? b) What about G6 (x0 )? Because the itinerary starts with an R and is not the itinerary of 1/2, x0 > 1 . Because the 2 seventh letter is L and the itinerary does not continue with that of 1/2, G6 (x) < 1 . 2 C1.4 Use a computer program to illustrate senstive dependence for the logistic map G(x) = 4x(1 x). Attempt to quantify how many interations are required to separate the points by 1/2 unit when the initial separation is 0.1,0.001, etc. Does the location of the initial pair matter? In the attached printout, we investigate the distance beetween the point 0.2 and the points 0.21, 0.201, and 0.2001. The iterations required to separate the points by at least a half were 9, 11, and 13 respectively. To check the dependence on location, we performed the same cacluations with 0.45 and 0.46, 0.451, and 0.4501. The iterations required to separate these points were 7, 10, and 20. Thus, location is important.