A Simple Proof of Sharkovsky's Theorem Author(s): Bau-Sen Du Source: The American Mathematical Monthly, Vol. 111, No. 7 (Aug.

- Sep., 2004), pp. 595-599 Published by: Mathematical Association of America Stable URL: http://www.jstor.org/stable/4145161 . Accessed: 31/05/2013 18:13
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 147.65.167.156 on Fri, 31 May 2013 18:13:49 PM All use subject to JSTOR Terms and Conditions

Herevsin Tois the "vertex sine"at To,the quantity givenby

vsin To =
- 2 cos Z ToTicos LTo T2cos LToT3-co2 V1

1 - cos2 TT2

cos2 LTo3.

It is not hardto provethis formulaandwe leave it to the reader. Fromthe powerseriesexpansions of the varioustermsin (7), we extractits "local" version(i.e., the well-knowncosine law for Euclidean tetrahedra):
A2 A

A + A - 2A1A2 CO ToT3-2A A3 cos LToT2 - 2A2A3 COS LToT1.

For an "octant" tetrahedron in which all dihedralangles at To (i.e., for a tetrahedron arerightangles)this reducesto the Pythagorean in three-space theorem [6]:
A2 O= = AA2 , +A2+ A2
AA. 3.

(8)

In a similarfashion,one can provethatin the case an "octant" tetrahespherical c-f dronit is truethat
A' < A +
A2

+ A.

In fact, it reduces(by settingAo/2 = w, A,/2 = x, A2/2 = y, and A3/2 = z) to the following:when0 < x, y, z, w < r, the conditions cos w = cos x cos y cos z + sinx sin y sin z, x + y + z > w, y + z + w > x, x + y + w > z, andx + z + w > y implythatw2 x2 + 2 + z2.
REFERENCES 1. M. Berger, Geometry,vol. 2, Springer-Verlag, New York, 1987. 2. , Encounterwith a geometer II, Notices Amer Math. Soc. 47 (2000) 326-340. 3. J. Cheeger and D. Ebin, Comparison Theoremsin Riemannian Geometry, North-Holland,Amsterdam, 1975. 4. D. Gromoll, W. Klingenberg, and W. Meyer, Riemannsche Geometrie im Grossen, Lecture Notes in Math., no. 55, Springer-Verlag, Berlin, 1968. 5. D. McDonell, privatecommunication. 6. D. Veljan,The 2500-year-old Pythagoreantheorem,Math. Magazine 73 (2000) 259-272. Departmentof Mathematics, Universityof Zagreb, Bijenieka 30, 10000 Zagreb, CROATIA dveljan@math.hr
<

A Simple Proof of Sharkovsky's Theorem

Bau-Sen Du
1. INTRODUCTION. Throughout this note, I is a compactinterval, andf : I I is a continuousmap. For each integer n > 1, let fn be definedby: f1 = f --and 2004] August-September
NOTES

595

This content downloaded from 147.65.167.156 on Fri, 31 May 2013 18:13:49 PM All use subject to JSTOR Terms and Conditions

: k > 0} the orbit of y (underf) and call y a periodicpoint of f with least periodm (or a period-m pointof f) if f"m (y) = y andf' (y) A y when0 < i < m. If f (y) = y, then we call y a fixedpoint of f. It is clearthatevery f of the type in questionhas fixed points. For discrete dynamicalsystems defined by iteratedinterval maps, one of the most remarkable results is Sharkovsky's theorem[5], [6]. It states that, if f has a then also a has period-mpoint, f period-npointpreciselywhenm -<n in the following Sharkovsky ordering:
o fn-1 when n > 2. For y in I, we call the set Of(y) = {fk(y) 3<5<7 -<23 3 2<2 .5<2
< 23 -< 22 -<2< 1.

f" = f

7 -

-<22 .3-<225<22-

It is well known(see [8]) thatthe sufficiencyof Sharkovsky's theoremcan be derived fromthe followingthreestatements: if has a (a) f periodicpointof leastperiodgreater than2, then f also has a periodicpointof least period2; (b) if f has a periodicpoint of odd periodm > 3, then f also has a periodicpoint of least periodn for every integern suchthatn > m + 1; (c) if f has a periodicpointof odd periodm > 3, then f also has periodicpointsof all evenperiods.The difficultyof provingthe sufficiency of Sharkovsky's theoremlies in proving(c), wheremostproofsinvolvethe structures of the so-called Stefancycles [1], [3], [7]. In this note, we give a unifiedproof of (b) and (c) thatdoes not involve Stefancycles. We also give a differentproof of (a) we end the paperwith a proofof Sharkovsky's [2], [8]. Forthe sakeof completeness, theorem. In provingSharkovsky's we needthe followingresult[3, p. 12]. theorem,
Lemma 1. Let k, m, n, and s be positive integers. Then thefollowing statements hold: (1) If y is a periodic point of f with least period m, then it is a periodic point of fn with least period m/ (m, n), where (m, n) is the greatest common divisor of m and n. (2) If y is a periodic point of fn with least period k, then it is a periodic point of f with least period kn/s, where s divides n and is relatively prime to k.

2. A PROOF OF (a). We needthe followingresult,whichcan also be used to show that f has no period-2point if and only if for each point c of I the iteratesf"(c) convergeto a fixedpointof f [3, p. 121].
Lemma 2. If c and d are points of I such that f (d) < c < d < f (c), then f has a periodic point of least period 2. Proof Write I = [a, b]. Let w = min{c < x < d : f (x) = x), and let v be a point in [c, w] with f(v) = d. Then,f2(v) = f(d) < c < v. If f has no fixedpointin [a, c], thenit fixes no pointof [a, v]. Since f2(a) > a, it follows thatf has a periodicpoint

with least period2 in [a, v]. If f has a fixedpointin [a, c], let t = max{a < x < c :
f(x) = x}. Then f has no fixed point in (t, v]. Let u be a point in [t, c] with f(u) = c. Then f2(u) = f(c) > d > u. Since f2(v) < v, we infer that f2(y) = y for some y

in [u, v]. Because f has no fixedpointin [u, v], y is a periodicpointof f with least period2. m
Proposition 3. If f has a periodic point of least period m larger than 2, then f also has a periodic point of least period 2. 596 OF AMERICA [Monthly 111 ASSOCIATION @ THEMATHEMATICAL

This content downloaded from 147.65.167.156 on Fri, 31 May 2013 18:13:49 PM All use subject to JSTOR Terms and Conditions

suchthatxs = max{x E P : x < f(x)}. It is clearthatxs+li By Lemma2, f has a periodicpointof leastperiod2.

Proof Let P = {xi : 1 <i < m}, with xl < x2 < '" < m, be a period-m orbit of f. Since xl < f(xi) and f(xm) < xm, there exists an " integer s satisfying 1 < s < m - 1

f(x,) and f(xsl)

< x,. i

3. A UNIFIEDPROOFOF (b) AND (c). If thereareclosed subintervals Jo, J1,..., of I with J, = Josuchthatf(Ji) D Ji+, for i = 0, 1, ..., n - 1, thenwe say Jn J,-1,
that JoJ1... J,-1 Jo is a cycle of length n. We require the following result [3, p. 7]: Lemma 4. If JoJ1J2 ... Jn-1J0 is a cycle of length n, then there exists a periodic , n - 1 and f" (y) = y. point y of f such that f' (y) belongs to Ji for i = 0, 1 ...

We now give a simpleunifiedproofof (b) and(c).

Proposition 5. If f has a periodic point of least period m with m > 3 and odd, then f has periodic points of all even periods. Furthermore,f has a periodic point of least period n for each integer n with n > m + 1. Proof Let P = {xi : 1 < i < m}, with xl < x2 < ... < X m, be a period-m orbit of f. Let x, = max{x E P : x < f (x)}. Then x,+l < f (xs) and f (xs+) < x,, so f has a fixed point z in [xs, xs+ ]. Since m is odd, for some integer t such that 1 < t < m - 1 andt : s the pointsf (xt) andf(x,+l) lie on oppositesides of z. Thusf ([xt, xt+1]) D

we assumethatx, < x,. If xs+1 < x,, the proofis similar. Let [xs,xs+l]. Forsimplicity, q be the smallestpositiveintegersuchthatfq (x,) < xt. Then2 < q < m - 1. Firstassumethatm = 3. Withoutloss of generality, we assumethat f(x1) = x2,

can applyLemma4 to the cycle Jo0JJ1 ... J1Jo of lengthn to obtaina period-npoint. if f has a period-3point,then f has periodicpointsof all periods.Now Accordingly, assume that m > 3. Since q is the smallestpositive integersuch that fq(x,) < xt, xt+1 < f'(xs) whenever1 < i < q - 1. If xt+l fq9- (Xs) < xs, Lemma4 appliesto the cycle

f (x2) =

x3, and f(x3)

x1. Let Jo = [x1, x2] and J1 = [x2, x3]. For any n > 2, we

x,),z][X,, [x fq-(X,)][fq-(Xs), fq-1(X,)] z][fq-I('

and establishesthe existence of a period-3point of f. If fq-I(xs) applyLemma4 to the cycle = Xs+l, we can

If k = min{1 < i < q - 1 : fq(xs) < < fq-1'(x) or < then so either xs+l fq-l(Xs), fk-l(xs) fi(x,)}, xt+l < fk-l(Xs) xl fk-l(x) X. If Xs+l < fk-I(Xs) < fq-l(Xs), we can invoke Lemma 4 for Weproceedassuming < that xs+1 fq-I(x,). the cycle

I][x,,Xs+1][z, xt+ Xs+1] xs+l][X,, to obtaina period-3pointof f. [z,

q-I(Xs)] fk-(X)][f-(Xs), fk-(Xs)][Z [fk-(X fq-1(Xs)][Z

to obtaina period-3pointof f. If x,+l < < xs (< z < fq-' (x)), we choose fk-(xs) u in [x,, xt+1]such that f(u) = z, pick w in [z, with f(w) = u, and let v in [fk-l(xs), z] be a point such that f(v) = w. fq-I(xs)] By applyingLemma4 to the cycle
August-September 2004] NOTES 597

This content downloaded from 147.65.167.156 on Fri, 31 May 2013 18:13:49 PM All use subject to JSTOR Terms and Conditions

[u, v][z, w][u, v] and, for every even integer n

>

4, to the cycle

U] [u,v]([z, w][v, z])(n-2)/2[z, W][u,

(here ([z, w][v, z])(n-2)/2 represents (n - 2)/2 copies of [z, w][v, z]) of length n, we conclude that f has periodic points of all even periods. On the other hand, let Ji = [z : f'(x,)] for i = 0, 1,..., q - 1, where [a : b] denotes the closed interval with a and b as endpoints. For any n m + 1, we appeal to Lemma 4 for the cycle of length n > . . JJo, where J = [x,, ], to confirm the existence of Jo J1 Xt+l]J J_-1 Jq-1 0 x,s+ a period-n point. [Xt, 4. A PROOF OF SHARKOVSKY'S THEOREM. We now combine (a), (b), (c), and Lemma 1 to prove Sharkovsky's theorem. Theorem 6 (Sharkovsky). Assume that f : I - I is a continuous map. If f has a period-m point, then f also has a period-n point precisely when m -< n in the Sharkovskyordering. Proof By (b) and (c), we have 3 -< 5 -< 7 < ...- < 2 3. If f has period-(2 - m) points with m > 3 and odd, then f2 has period-m points. By (b), f2 has period-(m + 2) points, which by Lemma 1(2) implies that f has either period-(m + 2) points or period-(2 - (m + 2)) points. If f has period-(m + 2) points, then by (b) f also has period-(2 - (m + 2)) points. In either case, f has period-(2 - (m + 2)) points. On the other hand, by (c) f2 has period-6 points and hence, by Lemma 1(2), f has period(22 - 3) points. Now if f has period-(2k - m) points with m > 3 and odd and if k > 2, then by Lemma 1(1) f2k-I has period-(2 - m) points. It follows from what we have just proved that f2k-I has period-(2 - (m + 2)) points and period-(22 . 3) points. In view of Lemma 1(2), f has period-(2k . (m + 2)) points and period-(2k+1 - 3) points. Furthermore, because f has period-(2k m) points, f2k has period-m points. By (b), f2k has period-2" points as long as 2" > ? m, so by Lemma 1(2) f has period-(2k+n) points for all integers n such that 2" > m. Finally, if f has period-2' points for some integer i > 2, then f2i-2 has period-4 points. As a result of (a), f2i-2 has period-2 points, ensuring that f has period-2i-' points. This proves the sufficiency of Sharkovsky's theorem. For the converse, it suffices to assume that I = [0, 1]. Let T(x) = 1 - 12x - 11 be the tent map on I. Then for any k > 1 the equation Tk(x) = x has exactly 2k distinct solutions in I. It follows that T has finitely many period-k orbits. Among these period-k orbits, let Pk be one with the smallest diameter max Pk - min Pk. For any x in I, let Tk(x) = min Pk if T(x) < min Pk, Tk(x) = max Pk if T(x) > max Pk, and Tk(x) = T(x) if min Pk < T(x) < max Pk. It is then easy to see that Tk has exactly one period-k orbit (i.e., Pk) and no period-j orbit for any j with j -< k in the Sharkovsky ordering (see also [1, pp. 32-34]). Now let Q3 be any period-3 orbit of T of minimal diameter.Then [min Q3, max Q3] contains finitely many period-6 orbits of T. If Q6 is one of smallest diameter, then [min Q6, max Q61contains finitely many period-12 orbits of T. We choose one, say Q 2, of minimal diameter and continue the process inductively. Let qo = sup{min Q2n.3 : n > 0) and q, = inf{max Q2n.3 : n > 0}. Let T,(x) = qo if T(x) < qo, T,(x) = ql if T(x) > q1, and T,(x) = T(x) if qo < T(x) < q1. Then it is easy to check that T, has periodic points of least period 2n for each n > 0, but has no periodic points of any other periods. This establishes the other U direction in Sharkovsky's theorem. 598

THEMATHEMATICAL OF AMERICA [Monthly 111 ASSOCIATION

This content downloaded from 147.65.167.156 on Fri, 31 May 2013 18:13:49 PM All use subject to JSTOR Terms and Conditions

Remark. Ourmethodcan also be used to provethat,if f has a periodicpointof odd periodm > 1 butno periodicpointsof odd periodstrictlybetween1 andm, thenany periodicorbitof oddperiodm mustbe a Stefanorbit(cf. [4]).
ACKNOWLEDGMENTS. I would like to thankM. Misiurewicz, A. N. Sharkovsky,and the referee for many constructivesuggestions that led to improvementsin this note.

REFERENCES 1. L. Alsed&,J. Llibre, and M. Misiurewicz, Combinatorial Dynamics and Entropy in Dimension One, 2nd ed., WorldScientific, Singapore, 2000. 107 (2000) 2. R. Barton and K. Bums, A simple special case of Sharkovskii's theorem, this MONTHLY 932-933. 3. L. S. Block and W. A. Coppel, Dynamics in One Dimension, LectureNotes in Math., no. 1513, SpringerVerlag, Berlin, 1992. 4. K. Burns, A note about Sharkovskii'stheorem, 2003 (preprint). 104 (1997) 846-847. 5. M. Misiurewicz, Remarkson Sharkovsky'stheorem, this MONTHLY 6. A. N. Sharkovsky,Coexistence of cycles of a continuous map of a line into itself, Ukrain. Mat. Zh. 16 (1964) 61-71; English translation,Internat.J. Bifur Chaos Appl. Sci. Engrg. 5 (1995) 1263-1273. 7. P. Stefan, A theorem of Sarkovskii on the existence of periodic orbits of continuous endomorphismsof the real line, Comm.Math. Phys. 54 (1977) 237-248. 8. P. D. Straffin,Jr.,Periodic points of continuous functions, Math. Mag. 51 (1978) 99-105. Instituteof Mathematics,Academia Sinica, Taipei 11529, Taiwan mabsdu@ccvax.sinica.edu.tw

The Hairy Ball Theorem via Sperner's Lemma

Tyler Jarvis and James Tanton
1. INTRODUCTION.It is well knownthatany continuous tangentvectorfield on the sphereS2 must,at some location,be zero. This resultis knownas the HairyBall for it can be loosely interpreted as follows: Theorem
It is impossible to comb all the hairs of a fuzzy ball so that (i) each hair lies tangent to the surface of the ball and (ii) the angles of the hairs vary continuously over the surface of the ball. (By this we mean that the angle between two hairs at positions p and q, say, can be made arbitrarily small by choosing q sufficiently close to p.) Any attempt to accomplish thisfeat must produce a cowlick.

We areassumingthateverypointof the ball's surfacesproutsa hair.It is a surprise to learn that this topologicalresult, like Brouwer'sfamous fixed point theorem[2, lemma. of Sperner's pp. 21-24], also follows froman application 2. SPERNER'S LEMMA. In 1928 EmanuelSpernerpresenteda simple, yet surlemmaabouttriangles[5]. We work with a slight prisinglypowerful,combinatorial of his result: original generalization
August-September 2004] NOTES 599

This content downloaded from 147.65.167.156 on Fri, 31 May 2013 18:13:49 PM All use subject to JSTOR Terms and Conditions