Sharkovskii's theorem

For some interval ${\displaystyle I\subset \mathbb {R} }$, suppose

${\displaystyle f:I\to I}$

is a continuous function. We say that the number x is a periodic point of period m if f ^m(x) = x (where f ^m denotes the composition of m copies of f) and having least period m if furthermore f ^k(x) ≠ x for all 0 < k < m. We are interested in the possible periods of periodic points of f. Consider the following ordering of the positive integers:

${\displaystyle {\begin{array}{cccccccc}3&5&7&9&11&\ldots &(2n+1)\cdot 2^{0}&\ldots \\3\cdot 2&5\cdot 2&7\cdot 2&9\cdot 2&11\cdot 2&\ldots &(2n+1)\cdot 2^{1}&\ldots \\3\cdot 2^{2}&5\cdot 2^{2}&7\cdot 2^{2}&9\cdot 2^{2}&11\cdot 2^{2}&\ldots &(2n+1)\cdot 2^{2}&\ldots \\3\cdot 2^{3}&5\cdot 2^{3}&7\cdot 2^{3}&9\cdot 2^{3}&11\cdot 2^{3}&\ldots &(2n+1)\cdot 2^{3}&\ldots \\&\vdots \\\ldots &2^{n}&\ldots &2^{4}&2^{3}&2^{2}&2&1\end{array}}}$

It consists of:

• the odd numbers in increasing order,
• 2 times the odds in increasing order,
• 4 times the odds in increasing order,
• 8 times the odds,
• etc.
• at the end we put the powers of two in decreasing order.

This ordering is a total order (every positive integer appears exactly once somewhere on this list), but not a well-order (e.g. there is no 'earliest' power of 2 in it).

Sharkovskii's theorem states that if f has a periodic point of least period m, and m precedes n in the above ordering, then f has also a periodic point of least period n.

As a consequence, we see that if f has only finitely many periodic points, then they must all have periods that are powers of two. Furthermore, if there is a periodic point of period three, then there are periodic points of all other periods.

Sharkovskii's theorem does not state that there are stable cycles of those periods, just that there are cycles of those periods. For systems such as the logistic map, the bifurcation diagram shows a range of parameter values for which apparently the only cycle has period 3. In fact, there must be cycles of all periods there, but they are not stable and therefore not visible on the computer-generated picture.

The assumption of continuity is important, as the discontinuous piecewise linear function ${\displaystyle f:[0,3)\to [0,3)}$ defined as:

${\displaystyle f:x\mapsto {\begin{cases}x+1&\mathrm {for\ } 0\leq x<2\\x-2&\mathrm {for\ } 2\leq x<3\end{cases}}}$

for which every value has period 3, would otherwise be a counterexample.

Similarly essential is the assumption of ${\displaystyle f}$ being defined on an interval – otherwise ${\displaystyle f:x\mapsto (1-x)^{-1}}$, which is defined on real numbers except the one: ${\displaystyle \mathbb {R} \setminus \{1\},}$ for which every non-zero value has period 3, would be a counterexample.

Sharkovskii also proved the converse theorem: every upper set of the above order is the set of periods for some continuous function from an interval to itself. In fact all such sets of periods are achieved by the family of functions ${\displaystyle T_{h}:[0,1]\to [0,1]}$, ${\displaystyle x\mapsto \min(h,1-2|x-1/2|)}$ for ${\displaystyle h\in [0,1]}$, except for the empty set of periods which is achieved by ${\displaystyle T:\mathbb {R} \to \mathbb {R} }$, ${\displaystyle x\mapsto x+1}$.[2][3]

Tien-Yien Li and James A. Yorke showed in 1975 that not only does the existence of a period-3 cycle imply the existence of cycles of all periods, but in addition it implies the existence of an uncountable infinitude of points that never map to any cycle (chaotic points)—a property known as period three implies chaos.[4]

Sharkovskii's theorem does not immediately apply to dynamical systems on other topological spaces. It is easy to find a circle map with periodic points of period 3 only: take a rotation by 120 degrees, for example. But some generalizations are possible, typically involving the mapping class group of the space minus a periodic orbit. For example, Peter Kloeden showed that Sharkovskii's theorem holds for triangular mappings, i.e., mappings for which the component f_i depends only on the first i components x₁,..., x_i.[5]

• Weisstein, Eric W. "Sharkovskys Theorem". MathWorld.
• "Sharkovskii's theorem". PlanetMath.
• Teschl, Gerald (2012). Ordinary Differential Equations and Dynamical Systems. Providence: American Mathematical Society. ISBN 978-0-8218-8328-0.
• Misiurewicz, Michal. "Remarks on Sharkovsky's Theorem". The American Mathematical Monthly,Vol. 104, No. 9 (Nov., 1997), pp. 846-847. Cite journal requires |journal= (help)
• Keith Burns and Boris Hasselblatt, The Sharkovsky theorem: a natural direct proof