Trinity University

Digital Commons @ Trinity

Mathematics Faculty Research Mathematics Department

12-1-2001

Periodic Points of the Family of Tent Maps

Julio R. Hasfura-Buenaga
Trinity University, jhasfura@trinity.edu

Phillip Lynch

Follow this and additional works at: http://digitalcommons.trinity.edu/math_faculty Repository Citation

Hasfura-Buenaga, Julio R. and Lynch, Phillip, "Periodic Points of the Family of Tent Maps" (2001). Mathematics Faculty Research. Paper 59. http://digitalcommons.trinity.edu/math_faculty/59

This Post-Print is brought to you for free and open access by the Mathematics Department at Digital Commons @ Trinity. It has been accepted for inclusion in Mathematics Faculty Research by an authorized administrator of Digital Commons @ Trinity. For more information, please contact jcostanz@trinity.edu.

PERIODIC POINTS OF THE FAMILY OF TENT MAPS

ROBERTO HASFURA-B. AND PHILLIP LYNCH

1. INTRODUCTION. Of interest in this article is the dynamical behavior of the one-parameter family of maps T (x) = (1/2 |x 1/2|), x R, > 0. For each such , the map T , which can be described piecewise by T (x) = x if x 1/2 , (1 x) if x > 1/2

is continuous, linear on each of the intervals (, 1/2] and [1/2, ) (with respective slopes and ,) and has the points (0, 0) and (1, 0) in its graph. The gure below, which shows the graph of the restriction of T2 to the unit interval I = [0, 1], illustrates why the maps of the family T are referred to as tent maps.

n (x)} , where T n denotes the nth iterate of T , If x R, the set {T n=0 is called the orbit of x. If T (x) = x then x is called a xed point, and k (x) = x for some positive integer k , then x (and its orbit) is called if T n (x) = x, for 1 n < k , x is said to have periodic or cyclic. If in addition T (prime) period k . The dynamical properties of a map T are to some extent determined by the nature of the orbits of points in its domain. For example, that the dynamical behavior of T when 0 < < 1 is docile can be inferred from the following facts concerning the orbits of T :

(i) T (0) = 0, so 0 is a xed point of T , (ii) T (1) = 0, so the point 1 is eventually xed, and n (x) 0 as n . (iii) for x = 0, 1, T (The dynamical behavior of T1 is even easier to understand: if a point is not already xed it is sent to one by an application of T1 .)
1

ROBERTO HASFURA-B. AND PHILLIP LYNCH

The behavior of T for all other values of is far more complicated as we now see. First, lets assume that > 2. In this case, by imitating the essential steps of the proof of Thm. ?? in [?], we can nd a Cantor set C [0, 1], invariant under T (that is, for which T (C ) = C ) with these properties:
n (x) as n ; (1) If x C then T (2) The restriction T |C of T to C is chaotic.

The term chaotic will be used throughout this paper in the sense of Devaney. Specically, a continuous dynamical system f : M M on the metric space (M, d) is called chaotic if: C1 It is topologically transitive, i.e. given nonempty open sets U, V M there is an n > 0 for which f n (U ) V is not empty; C2 The periodic points are dense in M ; and C3 f has sensitive dependence on initial conditions, i.e. That T is chaotic on C when > 2 is established by nding a symbolic model of T |C on which the three conditions above are easily veried (see [?].) As we will be concerned with periodic ponts, we mention here that this symbolic model also informs us that T has periodic points of all periods and allows us to count their number. Now assume, as we will for the rest of this paper, that 1 < 2. We begin with the simple observations that T (0) = 0, T (1) = 0, and n (x) (monotonically) as n . that if x < 0 or x > 1 then T Consequently, outside the interval (0, 1) the dynamical behavior of T is understood (and very simple.) Also worthy of notice is the fact that the interval I = [ 2 /2, /2] remains invariant under the action of T . Indeed, if 2 /2 x 1/2, then 2 /2 x < T (x) = x /2 because > 1; the condition 1/2 < x /2 forces 2 /2 T (x) = (1 x) /2. Furthermore, the points of (0, 1) \ I are eventually mapped into I : For 0 < x < 2 /2, let n0 be the smallest nonnegative integer for which n0 +1 x 2 /2. Then n0 x < 2 /2 < 1/2 implies that n0 +1 (x) < /2. On the other hand, if /2 < x < 1 then 2 /2 T 0 < T (x) = (1 x) < 2 /2. It follows from these considerations that to get a more complete picture of the dynamics of T for 1 < 2 it is necessary to shed light on the behavior of the restriction T |I . To this task we devote the rest of this paper. 2. 1 < 2. From now on we conne our attention to T |I , 1 < 2 which we will simply denote by T . We will further simplify our writing by nding models of these T s in a common interval as follows: Let I de1 note [0, 1]. We dene the map T : I I as T = h T h , where h : I I is given by h (x) = ( 2 )(1 x) + x. 2 2

PERIODIC POINTS OF THE FAMILY OF TENT MAPS

Because h is a homeomorphism, the relation between the maps T and T is that of conjugacy. We justify the introduction of the T s by reminding the reader that conjugate maps share dynamical properties. In particular, T is chaotic if and only if T is. The graph of a typical T is shown in gure 2.

We now claim that T (and thus T ) is chaotic if 2 < 2. To verify this claim all we need to check is that corresponding to any nondegenerate n interval J I there is an integer n for which T (J ) = I , for this condition obviously implies that T is topologically transitive and, as shown in [?], on intervals C1 implies both C2 and C3. Thus, we just prove the following: Lemma 1. If 2 < 2 and J I is a non degenerate interval then n there is a positive integer n for which T (J ) = I.

Proof. Set a = /( + 1), the xed point of T . We leave to the reader the verication that if J contains a in its interior then J is eventually mapped onto I . If a int(J ) there are several possibilities. First, we could have J = [a, b], a < b. Then T (J ) = [c, a], c < a, and if ( 1)/ [c, a] then |[c, a]| = |[a, b]|. (Here, | | denotes length.) Further applications of T m eventually yield m such that T (J ) = [d, a] and ( 1)/ [d, a]. Therem+1 m+2 m+3 fore, T (J ) [a, 1], T (J ) [0, a], T (J ) [2 , 1] and [2 , 1] has a in its interior because 2 < a = /( + 1) for > 2. But we are now in the situation contemplated in the previous paragraph. A second possibility is that J = [c, a], c < a. In this case, either T (J ) = [a, b], a < b (the previous case), or plainly T (J ) contains a in its interior. Again, we are in one of the previous two cases. Finally, it could happen that a J . 2 2 If in addition a T (J ) then |T (J )| 2 |J | > |J |. Therefore, repeated m application of T eventually yields m such that a T (J ) and again we get the desired conclusion by invoking the earlier cases.

ROBERTO HASFURA-B. AND PHILLIP LYNCH

Now assume that 1 < 2. It will be convenient to examine the map 2 T . The graph of one such map is sketched in the gure below.

It can be checked by straightforward computations that the interval Q1 = 2 2 [0, T (0)] = [0, 2 ] remains invariant under the action of T , i.e. that 2 2 2 if x [0, T (0)] then T (x) [0, T (0)]. Furthermore, if we dene the 2 new function F : I I by F = g 1 T g , where g : I Q1 is 2 2 the homeomorphism g (x) = (1 x)T (0) (so F and (T )|Q1 are conjugate) then, by another elementary computation, we see that F (a piecewise linear function) has the graph shown in Figure 4 below. It is apparent from this

graph that F = T . Therefore, we have proved the following: Theorem 1. (T )|Q1 is conjugate to T 2 . Using Theorem 1 along with some of our previous results we can draw 2 some conclusions. For example, if < 2 and 2 > 2 then T |Q1 , being conjugate to T 2 , is chaotic. Furthermore, for such , if we set Q2 = T (Q1 ) then Q1 and Q2 are disjoint intervals, Q2 = T (Q1 ) (so Q1 Q2 is T invariant), T is chaotic on Q1 Q2 , and every point in I \ Q1 Q2 is eventually mapped by T into Q1 Q2 . In other words, T |Q1 Q2 is a chaotic attractor. More generally, the following result can be obtained from judicious, repeated application of Theorem 1.
2

PERIODIC POINTS OF THE FAMILY OF TENT MAPS

Corollary 1. Assume that 1 < k < 2, 1 k n 1, and n > 2. Then there exist disjoint subintervals Q1 , Q2 . . . Q2n of I permuted by T such that Q = Q1 Q2 . . . Q2n is a chaotic attractor of T . That is, T |Q is chaotic, and every point in I \ Q is eventually mapped by T to Q. Similarly, if = 2 then Q1 = [0, a] where a = /( + 1) is the xed point of T , Q2 = T (Q1 ) = [a, 1] (so Q1 Q2 = I ), and T permutes Q1 and Q2 . Applying Theorem 1 once again we obtain: Corollary 2. T 2 is chaotic on I = [0, 1]. 3. PERIODIC POINTS. In the previous section we learnt that when 1 < 2 there is a set Q I on which T is chaotic. We know then, from C2 of the denition of chaos, that there is a dense (hence innite) subset of Q consisting of T -periodic points. However, in contrast to the case > 2, we dont have as yet any specic information about them. Myriad interesting questions can be asked regarding the periodic points of T , and we devote this section to answering some of them. Naturally, when considering such questions, the well-known theorem of Sarkovskii which establishes that the presence of certain periodic orbits forces the existence of certain other periodic orbits will be very useful. The precise statement of Sarkovskiis theorem is the following: Theorem 2 ([?]). Let (N, ) be the following transitive ordering of the positive integers: 3 5 7 . . . 2 3 2 5 3 7 . . . 22 3 22 5 22 7 . . . 2n 3 2n 5 2n 7 . . . . . . 2n 2n1 . . . 22 21 20 . Assume that J is an interval and that f : J J is continuous. If f has a periodic point of period k then it has a periodic point of period n for all k n. We will nd it convenient to make the following denition: k = inf{ > 1|T has a point of period k }. A few of the k s can be found by explicit algebraic computation. For example, 2 = 1, 3 = (1 + 5)/2, 4 = 1, 5 1.5129, and 6 = (1 + 5)/2. The relation observed between 3 and 6 is an instance of a general property which is consequence of our work above. In fact, several simple properties of k can be found easily using the geometry of T , Sarkovskiis theorem, and our previous results. For example, the reader can check that if > k then T has a periodic point of period k , and that k n implies n k , i.e. that the map : (N, ) (R, ) is order-reversing. Also, as we saw in the proof of Theorem 1, when 1 < 2 there is a set Q1 which intersects all the periodic orbits of T , for which Q2 = T (Q1 ) and Q1 intersect in a set 2) containing no more than one point, and for which (T |Q1 is conjugate to 2 T (on an appropriate set.) These facts yield the following: Proposition 1. (i) 2n+1 2; (ii) k = 2 2k Proof. (i) The properties of Q1 imply that T has no periodic points of odd period; (ii) by the conjugacy referred to before the statement of the

ROBERTO HASFURA-B. AND PHILLIP LYNCH

proposition, corresponding to a T2 -periodic point of period n = 1 there is 2 -period n. Since Q and its T -image Q intersect a point x in Q1 with T 1 2 at at most one point, x must have T -period 2n. We now describe a class of polynomials which will appear often in our discussion. We say that a polynomial alternates if it has the form p(x) = xa1 xa2 + xa3 . . . xai , where a1 > a2 > a3 > > ai 1, including the zero polynomial. We will denote arbitrary alternating polynomials of degree less than or equal to k by pk ; similarly, by qk we will denote arbitrary alternating polynomials of degree exactly k . To indicate that pk or qk have an even number of nonzero terms we will write p k or q k ; and to indicate that they have an odd number of nonzero terms we will write p k or q k . The k (x) can be relevance of alternating polynomials arises from the fact that T expressed in terms of them. The following result states this fact precisely.
k (x) = p k (x) = p Lemma 2. T k ( ) k x or T k ( ) + k x. Furthermore, in the expressions above we can make sure that deg(pk ) < k if x 1/2 and that deg(pk ) = k if x 1/2.

Proof. The proof is by induction. If k = 1 then T (x) is x or x depending on whether x 1/2 or x 1/2. In either case, the lemma is valid for k = 1. Now assume that the statement of the lemma is true for k = n 1, so
n1 n1 T (x) = p n2 ( ) n1 x or T (x) = p n2 ( ) + n1 x

if x 1/2, and
n1 n1 T (x) = q n1 ( ) n1 x or T (x) = q n1 ( ) + n1 x

if x 1/2. To verify the lemma for k = n we consider dierent cases. For n1 (x) 1/2 then T n (x) is T n1 (x), that is, example, if x 1/2 and T
n T (x) = ( pn2 ( ) n1 x) = p n2 ( ) n x = p n1 ( ) n x

or
n T (x) = (p n2 ( ) + n1 x) = p n2 ( ) + n x = p n1 ( ) + n x. n1 (x) 1/2 then T n (x) is T n1 (x), that is, Or if x 1/2 and T n T (x) = ( qn1 ( ) n1 x) = q n1 ( ) + n x = q n ( ) + n x

or
n T (x) = (q n1 ( ) + n1 x) = + q n1 ( ) n x = q n ( ) n x.

We see that in the two cases considered the statements of the lemma are veried. There are two more cases to consider, and we leave them to the diligent reader. In each case the statements of the lemma are again validated for k = n, and the proof is complete.

PERIODIC POINTS OF THE FAMILY OF TENT MAPS

Corollary 3. Assume that x is a point of period k . If x 1/2 then it can be represented in the form p k1 ( ) p k1 ( ) or k , k 1 +1 and if x 1/2 then it can be represented in the form x= q k ( ) q k ( ) or k . k 1 +1
k (x) = x. Now use the expressions for Proof. If x has T -period k then T k T (x) found in the preceding lemma and solve for x.

This last corollary raises hope that the problem of clarifying the nature of the periodic orbits of T may be aided by the study of expressions of the form p k ( ) p k ( ) and k . k 1 +1 Such hope is further strenghtened by the following observations: k ( ) is closed under the following (a) Given k and , the (nite) set p k 1 operation R: p k ( ) p k ( ) k if deg( pk ) < k , and k 1 1 p k ( ) p k ( ) k if deg( pk ) = k . k 1 1
k ( ) (b) Given k and , the (nite) set p is closed under the following k +1 operation R: p k ( ) p k ( ) k if deg( pk ) < k , and k + 1 +1 p k ( ) p k ( ) k if deg( pk ) = k . k + 1 +1 (Notice that the description of the action of T on x, whose form depends on whether x 1/2 or x 1/2, is identical to the description of the given operation R, only now the dependence is on deg(pk ).) k ( ) The operation R just introduced partitions each of the sets p and k 1 p k ( ) k +1 p k ( ) k 1

into cycles (or orbits.) We will say that one of these cycles in (resp. in
p k ( ) k +1

) satises the 1/2-condition for if

p k ( ) p k ( ) 1/2 and deg( pk ) = k if k 1/2 k 1 1 p k ( ) p k ( ) (resp. deg( pk ) < k if k 1/2 and deg( pk ) = k if k 1/2) +1 +1 for all the elements of the cycle. (The ambiguity concerning the value 1/2 is irrelevant.) We can now state the following important result. deg( pk ) < k if

ROBERTO HASFURA-B. AND PHILLIP LYNCH

k ( ) k ( ) Theorem 3. Given and k , if x = p or x = p and the R-cycle of k +1 k 1 k (x) = x. the rational expression satises the 1/2-condition then T k ( ) Proof. Suppose x = p . We study the orbit of x under T by studying k +1 the orbit of the rational expression under the operation R; this is possible since both actions are identical given that the orbit of the rational expression satises the 1/2-condition. To simplify notation, we consider only the coecients of the polynomial p k ( ), which we present as an element of S k , where S = {1, 0, 1}. Thus, if v = (v1 , v2 , . . . , vk ) is the element of S k containing the coecients of p k ( ) then the action R : S k S k is given by

R(v )i = vi1 for i > 1, and R(v )1 = 0 if vk = 0, and R(v )i = vi1 for i > 1, and R(v )1 = 1 if vk = 0. That is, R(v ) = (|vk |, (1)|vk | v1 , (1)|vk | v2 , . . . , (1)|vk | vk1 ) and, continuing the same way Rk (v ) = (|v1 |, (1)|v1 | |v2 |, (1)|v1 |+|v2 | |v3 |, . . . , (1)|v1 |+...|vk1 | |vk |). Now, for any polynomial of the form p k , vi = (1)X |vi |, where X = i1 j =1 |vj | because for every term with coecient 1 there is an even number of terms of smaller degree in the polynomial, while for every term with coecient 1 there is an odd number of terms of smaller degree in the k (x) = x. polynomial. Therefore, since Rk (v ) = v we must have T k ( ) is very similar. We just use the The proof for the case when x = p k 1 appropriate form for R, namely R(v )i = vi1 for i > 1, and R(v )1 = 0 if vk = 0, and R(v )i = vi1 for i > 1, and R(v )1 = 1 if vk = 0, and the specic properties of the polynomials p k . Suppose now that k is an odd number 5, and consider the polynomial s1 ( ) = k1 k2 + k3 . . . . This polynomial is of the form p k ( ). s 1 ( ) Also consider the R-cycle of , consisting, in addition to that rational k 1
s 2 ( ) s 3 ( ) expression itself, of k 1 , k 1 , . . . , k k 1 s2 ( ) = + k 2 . . . 2 , k k 1 s3 ( ) = + . . . + 3 , . . . s i ( ) , k 1

...

sk1 ( ) , k 1

and

s k ( ) . k 1

Here

si ( ) = k k1 + . . . + (1)(i+1) i + (1)i i2 + . . . , . . .

PERIODIC POINTS OF THE FAMILY OF TENT MAPS

sk1 ( ) = k k1 + k3 . . . , sk ( ) = k k2 + k3 . . . .
1 1/2 for some > 1 then the R-cycle above satises Lemma 3. If k k 1 the 1/2-condition for that .

( )

Proof. First consider those si having degree k . If 2 i k 2, we would like that sk1 ( ) si ( ). That this is so follows from the fact that si ( ) sk1 ( ) = ( 1)( k3 k4 + . . . +(1)(i+1) i+1 ) > 0 for all > 1. Thus, sk1 ( ) s i ( ) k 1 k 2 > 1/2 forces k 1 > 1/2. Likewise, sk ( ) sk 1 ( ) = k 1 0 for > 1 so
s k ( ) k 1

> 1/2 as well.

s ( )

s 1 ( ) 1 Finally, we need to show that k 1/2 implies that k 1 1/2. By k 1 cross-multiplying and subtracting in the last inequality, we reformulate our desideratum as k 2 k1 + 2 k2 . . . + 2 1 0. But, again crossmultiplying and subtracting, the rst inequality can be written as k 2 k1 + 2 k3 2 k4 + . . . 2 + 1 > 0 and this inequality is easily seen to imply the previous one when > 1.

It is now a consequence of Theorem 3 and Lemma 3 that if for some > 1, s 1 ( ) k (x). One such is, of course, the largest real x= k 1/2, then x = T k 1 root of fk (x) = xk 2xk1 + 2xk3 . . . 2x + 1 which will henceforth be 1 ( k ) k denoted by k . Thus, x = s k 1 satises x = T k (x). We have not excluded k n for some 1 n < k , i.e. that x have the a priori possibility that x = T k (prime) period < k . If this indeed happened then for some 2 i k we would have s1 ( k ) = si ( k ). However, these two polynomials (and in fact any pair sj , sl , j = l) can have the same value at only nitely many points. Therefore we can select k arbitrarily close to, but larger than k , i.e. in [ k , k + ) for arbitrarily small , so that fk (k ) 1/2 (fk is nondecreasing in a neighborhood of its largest real root), and so that si (k ) = sj (l ), j = l. ( k ) In this case x = s1k is Tk -periodic of (prime) period k . From this the k 1 inequality k k follows. This inequality is useful because we will be able to show that limn 2n+1 = 2. Lemma 4. (i) fk ( 2) = 2 2 3 for all odd k ; and (2) limk fk (x) = for x > 2.
1 Proof. (i) fk (x) = xk 2xk1 + 1 + (1 x )(xk3 + xk5 + . . . + x2 ). Thus k k 2 x x 2 1 x x 3 2 fk (x) = xk 2xk1 + 1 + 2x x2 +1 +1 = x+1 (x x 2x + 2) + 1+x . Now 2 = 2 2 3. (ii) When x > 2 a simple we compute fk ( 2) = 0 + 1 2+1

calculation shows that x3 x2 2x + 2 > 0. Thus, fk (x) = x x+1 for some > 0, implying that limn f2n+1 (x) = . Theorem 4. limn 2n+1 = 2.

k2

10

ROBERTO HASFURA-B. AND PHILLIP LYNCH

Proof. By the preceding lemma, given any once n is large enough we > 0, will have f2 ( 2 + ) > 1. Since f ( 2) = 2 2 3 < 0 for such n, we n+1 2n+1 must have 2 < 2n+1 < ( 2 + ). Now let decrease to 0. Corollary 4. limn 2n+1 = 2. Proof. From Proposition 1(i) and the comment preceding Lemma 4 we have 2 2n+1 2n+1 . Now use Theorem 4. Thus, T2 has, besides the xed point, periodic points of all even periods and none of odd period. Also, m1 Corollary 5. (i) limn (2n+1)2m = 22 ; (ii) limm (2n+1)2m = 1. Corollary 6. For all m, 2m = 1. Proof. This follows from part (ii) of the Corollary 5 and from the previously mentioned fact that : (N, ) (R, ) is order-reversing. References
[D] R. L. Devaney, An Introduction to Chaotic Dynamical Systems, Second Edition, Addison Wesley, Menlo Park, CA, 1989. [VB] M. Vellekoop and R. Berglund, On Intervals, Transitivity = Chaos, Amer. Math. Monthly, 101, No. 4 (1994), 353-355. [S] A. N. Sarkovskii, Coexistence of cycles of a continuous mapping of a line into itself, Ukr. Mat. Z., 16, 61-71, 1964. Department of Mathematics, Trinity University, San Antonio, TX 78212 E-mail address : jhasfura@trinity.edu Department of Mathematics, University of Washington, Seattle, WA 98195 E-mail address : lynch@math.washington.edu