First, a quick short Re-Cap ...
PIECEWISE LINEAR ONE-DIMENSIONAL MAPS
First example : the TENT map
xn+1 = F (xn ) = 1 − 2|xn − 1/2|
0.8
n+1
0.6
at time
0.4
x
0.2
0
0 0.2 0.4 0.6 0.8 1
x at time n
I Initial conditions that are negative move off to −∞
Doubling their distance from the origin on each iterate
I If x0 > 1, then x1 < 0 and the subsequent orbit moves off to
−∞
I For x ∈ [0, 1], we have 0 ≤ 1 − 2|x − 1/2| ≤ 1
So all orbits are bounded in the interval
Time evolution of the state x
0.8
n
at time
0.6
0.4
x
0.2
0 20 40 60 80 100
Iterate number n
Map has 2 actions :
I Interval uniformly stretched to twice its original length
I Stretched interval is folded in half – so that the folded line
segment is now contained in the original interval
STRETCHING and FOLDING :
Following a point under the map —
I Stretching leads to exponential divergence of nearby
trajectories (by a factor of two on each iterate)
I Folding : keeps orbit bounded
Folding makes it non-invertible : since it results in 2 different
values of xn mapping the same xn+1
To demonstrate sensitive dependence on initial conditions:
Compose the map m times with itself to obtain F m :
F m (x) = F (F m−1 (x)) = F (F (F (. . . (F (x))) . . . ))
with F 1 (x) = F
Thus xn+m (x) = F m (xn )
For instance, xn+2 = F (F (xn )) = F 2 (xn )
So, in case of TENT MAP:
xn+2 = F (F (xn )) = 1 − 2| (1 − 2|xn − 1/2|) −1/2|
1
I If xn = 0, 2, 1
Then 2 applications of F yield xn+2 = 0
I If xn = 14 , 3
4
Then 2 applications of F yield xn+2 = 1
Linear in between
x-axis : xn
y-axis : xn+2 = F (F (xn )) = F 2 (xn )
0.8
n+2
0.6
at time
0.4
x
0.2
0
0 0.2 0.4 0.6 0.8 1
x at time n
x-axis – Initial state xn
y-axis – mth iterate starting from initial state xn , i.e. state x at
time n + m : xn+m = F m (xn )
0.9
0.8
0.7
0.6
X at time n+m
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
X at time n
Sixth Iterate : i.e. m = 6
Initial condition lies within 21m of some point, the xm can lie
anywhere in the interval [0, 1]
Absolutely no knowledge of the location of the future points after
times > m even though you know the initial point with reasonable
precision
Exponential Sensitivity of chaotic orbits to small changes in initial
conditions
Does God Play Dice?
Key idea:
The present determines the future, but approximate present does
not approximately determine the future
Irregular motion of a deterministic dynamical system that is so
sensitive to initial conditions that they are impossible to predict in
the long term with anything less than an infinite and perfect
representation of analog values
Another simple map:
xn+1 = 2xn mod 1
Can be regarded as a map on a circle
Modulo 1 : makes x like an angle variable : x increasing from 0 to
1 corresponds to one circuit around the circle
Action of the map: Stretch - Twist - Fold operation
Chaotic separation of nearby points (by a factor of 2) on each
iterate of this map
x-axis – Initial state xn
y-axis – Subsequent iterate starting from initial state xn , i.e.
state x at time n + 1 : xn+1 = F (xn ) = 2x mod 1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
Alternate way of viewing: Bernoulli Shift
Say we represent the initial condition x0 as a binary decimal
∞
X
x0 = 0.a1 a2 a3 · · · ≡ 2−j aj
j=1
where each digit aj is either 0 or 1
Then next iterate is obtained by setting the first digit to 0 and
moving the decimal point to the right :
x1 = 0.a2 a3 a4 . . .
x2 = 0.a3 a4 a5 . . .
and so on
So digits initially far right – having only very slight influence on the
initial value of x eventually become first digit
Small change in initial condition
Say : a10 is changed from 0 to 1,
i.e. a change in x0 by 2−10
At time n = 9 makes huge change in xn
Periodic orbit of a map
If the orbit successively cycles through p distinct points:
x0 , x1 , . . . xp−1
Such that xi 6= xj if i 6= j
Graphical interpretation of Period-1 for Tent map
Fixed Point (i.e. p = 1)
xn+1 = F (xn ) = xn
i.e.
xn = 1 − 2|xn − 1/2| = x?
Orbit has one point: x? , x? , x? . . .
1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
x-axis : xn
y-axis (Solid Line) : xn+1 = F (xn ) = 1 − 2|xn − 1/2|
y-axis (Dashed Line) : xn+1 = xn (450 line)
Intersections of the two lines : Period 1 orbit solutions :
x? = 0, x? = 2/3
I Fixed Point Solution x ≤ 1/2:
xn+1 = 2xn = xn
Only solution : x? = 0
I Fixed Point Solution x > 1/2:
xn+1 = 2 − 2xn = xn
Only solution : x? = 2/3
Period 2 cycle
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
y-axis (Solid blue Line) : xn+1 = F (xn )
y-axis (Solid magenta Line) : xn+2 = F 2 (xn )
y-axis (Dashed Line) : 450 diagonal line
Intersections of the solid and dashed lines : Period 2 orbit solutions
Notice that period 1 solutions are also period 2 solutions
Since, if F (xn ) = xn , then F 2 (xn ) = F (xn ) = xn
So two intersection points (0 and 2/3) will match for both xn+1
and xn+2 .
Period 2 Solution: xn+2 = xn
Four possibilities xn → xn+1 → xn+2 ≡ xn :
I xn < 1/2 and xn+1 < 1/2
xn+2 = 2xn+1 = 2 × 2xn = xn
Solution : 0
I xn < 1/2 and xn+1 > 1/2
xn+2 = 2 − 2xn+1 = 2 − 2(2xn ) = xn
Solution : 2/5
I xn > 1/2 and xn+1 < 1/2
xn+2 = 2xn+1 = 2(2 − 2xn ) = xn
Solution : 4/5
I xn > 1/2 and xn+1 > 1/2
xn+2 = 2 − 2xn+1 = 2 − 2(2 − 2xn ) = xn
Solution : 2/3
Period 3 cycle
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
y-axis (Solid blue Line) : xn+1 = F (xn )
y-axis (Solid magenta Line) : xn+3 = F 3 (xn )
y-axis (Dashed Line) : 450 diagonal line
Intersections of the solid and dashed lines : Period 3 orbit solutions
Graphical interpretation of Fixed Point for Shift Map
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
x-axis : xn
y-axis (Solid Line) : xn+1 = F (xn ) = 2x mod 1
y-axis (Dashed Line) : 450 line
Intersections of the two lines : Period 1 orbit solutions
Period 1 orbit of Shift Map F (x) = 2x mod 1
Fixed Point: xn+1 = xn
I x0 = 0.0000 . . .
x1 = 0.0000 · · · = x0
I x0 = 0.1111 . . .
x1 = 0.1111 · · · = x0
Period 2 orbit
2
If x0 = 3 = 0.10101010 . . .
1
x1 = 0.01010101 · · · = 3
2
x2 = 0.10101010 · · · = 3
Period-2 cycle (x0 , x1 ), namely 23 , 1 2 1
3, 3, 3 ...
General period of order p arises from initial conditions of the form:
0.a1 a2 a3 . . . ap a1 a2 . . .
Graphical interpretation of Period 2 orbit for Shift Map
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
y-axis (Solid magenta Line) : xn+2 = F 2 (xn )
y-axis (Dashed Line) : 450 line
Intersections of the lines : Period 2 orbit solutions
Two intersections common to Period 1
We ask : How many different initial conditions are there that
return to themselves after p iterations ?
There are 2p distinct sequences
2p intersections with xn+p = xn
A point on the period p orbit is a fixed point of the composed map
F p as
xn = F p (xn )
Two of these : fixed points 0 and 1
For p = 2 : 2p = 22 = 4
i.e. there are 4 intersections of the diagonal line xn+2 = xn with
F 2 (xn )
1 2
Intersections at 0, 3, 3, 1
If p is a prime number then all these 2p − 2 points lie on orbits of
period p
So number of distinct periodic orbits:
2p − 2
Np =
p
This number gets large rapidly:
e.g.
For p = 11 : N11 = 186
For p = 17 : N17 = 7710
If p is not a prime number :
Has integer factors p1 , p2 , . . . (p = pn1 1 pn2 2 . . . )
Then some of these 2p − 2 points will lie on orbits of lower period
p1 , p2 . . .
e.g. For p = 4, 2p − 2 = 14
Out of which 2 points belong to p = 2
2p
But here too for large p : Np ∼ p
Question: What is the stability of these orbits?
