-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathiterating.tex
145 lines (133 loc) · 5.16 KB
/
iterating.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
\documentclass[a4paper]{article}
\usepackage[english]{babel}
\usepackage[utf8x]{inputenc}
\usepackage{amsmath}
\usepackage{graphicx}
\usepackage{hyperref}
\usepackage[colorinlistoftodos]{todonotes}
\title{Playing with function iteration}
\author{Slavomir Kaslev}
\begin{document}
\maketitle
\section{General problem}
The equations $f(x) = x + f(x^2 + x^3)$ and $f(x) = x + f(\frac{1}{1-x})$ are instances of the more general equation
\begin{equation}
f(x) = x + f(g(x))\label{eq:eq1}
\end{equation}
We want to find a function $f(x)$ such that \eqref{eq:eq1} holds for all $x$, where the function $g(x)$ is known. Telescoping the equation gives
$$f(x) = x + g(x) + g(g(x)) + g(g(g(x))) + \dots$$
Let's use the following notation for iterating a function
\begin{align*}
g^{[0]}(x) &= x\\
g^{[1]}(x) &= g(x)\\
g^{[n]}(x) &= g(g^{[n-1]}(x))
\end{align*}
to rewrite equation \eqref{eq:eq1} as
\begin{equation}
f(x) = \sum_{n=0}^{\infty}{g^{[n]}(x)}\label{eq:eq2}
\end{equation}
We've reduced the problem from solving equation \eqref{eq:eq1} to calculating the infinite sum \eqref{eq:eq2}.
\section{Power series sleight of hand}
The next step is instead of solving equation \eqref{eq:eq1} we'll change the problem by inserting a small parameter $\epsilon$
\begin{equation}
f(x) = x + \epsilon f(g(x))\label{eq:eq3}
\end{equation}
Repeating the steps above, we telescope \eqref{eq:eq3}
$$f(x) = x + \epsilon g(x) + \epsilon^2 g(g(x)) + \epsilon^3 g(g(g(x))) + \dots$$
and finally we get $f(x)$ as power series of $\epsilon$
\begin{equation}\label{eq4}
f(x) = \sum_{n=0}^{\infty}{g^{[n]}(x)} \epsilon^n
\end{equation}
Now we have two procedures for computing $f(x)$ defined by equations \eqref{eq:eq2} and \eqref{eq4}. Of course, the two procedures will agree only if we finally set $\epsilon$ to $1$.
\section{Special case $g(x) = \frac{1}{1-x}$}
Equation \eqref{eq:eq3} takes the form
\begin{equation}\label{eq5}
f(x) = x + \epsilon f(\frac{1}{1-x})
\end{equation}
Note that the function $g^{[n]}(x)$ is periodic in $n$ with period 3. That is
\begin{align*}
g^{[0]}(x) &= x\\
g^{[1]}(x) &= \frac{1}{1-x}\\
g^{[2]}(x) &= 1 - \frac{1}{x}\\
g^{[3]}(x) &= x
\end{align*}
So the series in \eqref{eq4} have the form
\begin{equation}\label{eq6}
f(x) = x + \epsilon \frac{1}{1-x} + \epsilon^2 (1 - \frac{1}{x}) + \epsilon^3 x + \epsilon^4 \frac{1}{1-x} + \dots
\end{equation}
We'll sum this series using a generic summation procedure $S$ with the properties
\begin{subequations}
\begin{align}
S(a_0 + a_1 + a_2 + \dots) &= a_0 + S(a_1 + a_2 + a_3 + \dots)\label{eq:sum1}\\
S(\sum{\alpha a_n} + \sum{\beta b_n}) &= \alpha S(\sum{a_n}) + \beta S(\sum{b_n})\label{eq:sum2}
\end{align}
\end{subequations}
To simplify the calculation we'll substitute
\begin{equation*}
s = f(x)
\qquad
a = x
\qquad
b = \frac{1}{1-x}
\qquad
c = 1 - \frac{1}{x}
\end{equation*}
in \eqref{eq6} to obtain
$$s = S(a + b \epsilon + c \epsilon^2 + a \epsilon^3 + b \epsilon^4 + c \epsilon^5 + \dots)$$
Next we'll pull the first term out of the sum two times by \eqref{eq:sum1} and bring $\epsilon$ outside the sum by \eqref{eq:sum2}
\begin{align*}
s = S&(a + b \epsilon + c \epsilon^2 + a \epsilon^3 + b \epsilon^4 + c \epsilon^5 + \dots)\\
s = a + \epsilon S&(b + c \epsilon + a \epsilon^2 + b\epsilon^3 + c\epsilon^4 + a \epsilon^5 + \dots)\\
s = a + b \epsilon + \epsilon^2 S&(c + a \epsilon + b\epsilon^2 + c\epsilon^3 + a \epsilon^4 + b \epsilon^5+ \dots)\\
\end{align*}
If we multiply the first equation by $\epsilon^2$, the second by $\epsilon$ then add them all up, the result is
$$(1 + \epsilon + \epsilon^2)s = (1+\epsilon)a + \epsilon b + \epsilon^2 (a+b+c) S(1 + \epsilon + \epsilon^2 + \epsilon^3 + \dots)$$
Notice that $S(1 + \epsilon + \epsilon^2 + \epsilon^3 + \dots) = \frac{1}{1-\epsilon}$ and now we have an explicit formula for $s$
$$s=\frac{(1+\epsilon) a + \epsilon b}{1+\epsilon+\epsilon^2} + \frac{\epsilon^2}{1-\epsilon^3}(a+b+c)$$
Therefore the solution of \eqref{eq5} is
\begin{equation}
f(x) = \frac{(1+\epsilon)x + \epsilon\frac{1}{1-x}}{1+\epsilon+\epsilon^2} + \frac{\epsilon^2}{1-\epsilon^3}\frac{x^3-3x+1}{x(x-1)}\label{eq:soleps}
\end{equation}
We're interested in the solution of this equation
\begin{equation}
f(x) = x + f(\frac{1}{1-x})\label{eq9}
\end{equation}
which can be derived from \eqref{eq:soleps} by seting $\epsilon$ to $1$.
Notice that when $\epsilon=1$, the second term in the equation becomes infinite unless $x^3-3x+1=0$.
We conclude that the solution $f(x)$ of equation \eqref{eq9} is infinite everywhere except for a finite set of points given by the roots of the equation
$$x^3-3x+1=0$$
namely
\begin{equation*}
x_1 \approx -1.87938524
\qquad
x_2 \approx 0.34729636
\qquad
x_3 \approx 1.53208889
\end{equation*}
where $f(x)$ has an explicit formula
$$f(x)=\frac{1}{3}\;\frac{1 + 2x - 2x^2}{1-x}$$
and can be evalutated directly
\begin{equation*}
f(x_1) \approx -1.13715804
\qquad
f(x_2) \approx 0.7422272
\qquad
f(x_3) \approx 0.39493084
\end{equation*}
Notice that iterating $g(x)$ over the numbers $x_1,x_2,x_3$ forms a cycle
\begin{equation*}
g(x_1) = x_2
\qquad
g(x_2) = x_3
\qquad
g(x_3) = x_1
\end{equation*}
and
\begin{equation*}
\sum_{k=1}^{3}{x_k} = 0
\qquad
\sum_{k=1}^{3}{f(x_k)} = 0
\qquad
\sum_{k=1}^{3}{g(x_k)} = 0
\end{equation*}
\end{document}