5

Theorem 1
If $g \in [a,b]$ and $g(x) \in [a,b] \forall x \in [a,b]$, then $g$ has a fixed point in $[a,b].$
If in addition, $g'(x)$ exists on $(a,b)$ and a positive constant $k < 1$ exists with $$|g'(x)| \leq k, \text{ for all } \in (a, b)$$ then the fixed point in $[a,b] is unique.

Fixed-point Theorem Let $g \in C[a,b]$ be such that $g(x) \in [a,b]$, for all $x$ in $[a,b]$. Suppose, in addition, that $g'$ exists on $(a,b)$ and that a constant $0 < k < 1$ exists with $$|g'(x)| \leq k, \text{ for all } x \in (a, b)$$ Then, for any number $p_0$ in $[a,b]$, the sequence defined by $$p_n = g(p_{n-1}), n \geq 1$$ converges to the unique fixed-point in $[a,b]$

These are two theorems that I have learned, and I'm having a hard time with this problem:

Given a function $f(x)$, how can we find the interval $[a,b]$ on which fixed-point iteration will converge?

Besides guess and check, I couldn't find any other way to solve this problem. I tried to link the above theorems, but it involves two variables, so I have a feeling it can't be solved algebraically. I wonder is there a general way to find the interval of convergence rather trial and error? Thank you.

roxrook
  • 12,081
  • 2
    Well, any interval stable by $f$ and such that $|f'|\le k$ on this interval with $k<1$ (that is what the theorem tells you, no?). If $f(x)=(x+3/x)/2$ for example, $f'(x)=(1-3/x^2)/2$ hence any closed interval included in $(1,+\infty)$ will do. – Did Oct 03 '11 at 07:32
  • 1
    In general, the "basin of convergence" for an iteration method like Newton-Raphson has a fractal boundary. See this for instance. – J. M. ain't a mathematician Oct 03 '11 at 14:06

2 Answers2

4

Suppose the continuous function $f: {\mathbb R} \to {\mathbb R}$ has an attracting fixed point $p$. Its immediate basin of attraction (i.e. the maximal interval $J$ such that iteration of $f$ starting at any point in $J$ converges to $p$) must be an open interval $(a,b)$ (infinite intervals allowed). If $a$ is finite, $f(a)$ must be either $a$ or $b$, and similarly for $f(b)$. The possibilities are as follows:

1) $a = -\infty$, $b = \infty$

2) One of $a$ and $b$ is a (non-attracting) fixed point, the other is infinite.

3) Both $a$ and $b$ are (non-attracting) fixed points.

4) $a$ is a (non-attracting) fixed point and $f(b) = a$

5) $b$ is a (non-attracting) fixed point and $f(a) = b$

6) $a$ and $b$ form a 2-cycle: $f(a) = b$, $f(b) = a$.

To tell which case you are in, find the non-attracting fixed points of $f$, the points mapped to those, and the 2-cycles. Note that $(a,b)$ can't contain any of those.

Robert Israel
  • 448,999
3

The iteration converges for any starting point in an interval stable by $f$ and such that $|f'|\leqslant k$ on this interval, with $k<1$ (this is a mere rephrasing of the theorem). If $f(x)=(x+3/x)/2$ for example, $f'(x)=(1-3/x^2)/2$ hence any closed interval included in $(1,+\infty)$ will do.

Did
  • 279,727
  • 1
    Your condition is sufficient but not necessary. In your example, the actual basin of attraction is $(0,\infty)$. – Robert Israel Oct 03 '11 at 18:18
  • Sure. The interval $(1,+\infty)$ is the one the general theorem the OP is interested in provides. By the way, what you call my condition is hardly mine (as I indicate in my answer). – Did Oct 03 '11 at 18:42