Newton Convergence

In this post, we establish that Newton’s method works under sufficiently mild conditions.

Let f : [-1, 1] \to \mathbb R have a continuous first derivative f' > 0 and suppose

f(-1) < 0 < f(1).

Problem 1. Show that there exists a unique c \in (-1, 1) such that f(c) = 0.

(Click for Solution)

Solution. By the intermediate value theorem, there exists c \in (-1, 1) such that f(c). For uniqueness, fix any root c' \in (-1,1). If c' \neq c, then by Rolle’s theorem, there exists c'' between c, c' such that 0 < f'(c') = 0, a contradiction. Hence, c = c', establishing uniqueness.

Problem 2. For any t \in (c, 1), denote the x-intercept of the tangent to y = f(x) at (t, f(t)) by t_0. Show that t_0 < t.

(Click for Solution)

Solution. Firstly, if f(t) \leq 0, then the intermediate value theorem gives us a second root c' \in [t, 1), contradicting the uniqueness of c in Problem 1. Therefore, f(t) > 0.

The tangent to y = f(x) at (t, f(t)) has equation

y = f(t) + f'(t) \cdot (x-t).

Therefore, 0 = f(t) + f'(t) \cdot (t_0-t). Doing some algebruh,

\displaystyle t_0 = t- \frac{ f(t) }{ f'(t) } < t,

since f(t) > 0 and f'(t) > 0.

Now furthermore suppose that f' is non-decreasing.

Problem 3. Show that t_0 > c, as defined in Problems 1 and 2.

(Click for Solution)

Solution. We observe that if t_0 < c, then f(t_0) < 0. Taking the contrapositive, it suffices to check that f(t_0) > 0. To that end, define the function

g(x) := f(x) - (f(t) + f'(t) \cdot (x-t)).

By definition, f(t_0) = g(t_0), so it suffices to prove that g \geq 0. To that end, we use the mean value theorem to furnish \xi between t and x such that

f(x) = f(t) + f'(\xi) \cdot (x-t).

Substituting and subtracting,

g(x) = (f'(\xi) - f'(t)) \cdot (x-t).

If x < t, then \xi < t so that f'(\xi) \leq f'(t). Hence, g(x) \geq 0. In particular, g(t_0) \geq 0, so that f(t_0) \geq 0. Since f(t_0) \neq 0, we must have f(t_0) > 0. Therefore, t_0 > c.

Now furthermore suppose that f has a second derivative such that f'' > 0. Define the sequence \{x_n\} by

\displaystyle x_0 := 1,\quad x_{n+1} := x_n - \frac{ f(x_n) }{ f'(x_n) }.

Problem 4. Show that x_n \to c.

(Click for Solution)

Solution. By Problem 2, x_{n+1} < x_n. By Problem 3, x_{n+1} > c. Therefore, \{x_n\} is decreasing and bounded below. By the monotone convergence theorem, x_n \to \xi for some \xi \in [-1, 1]. Thanks to convergence and continuity,

\begin{aligned} \xi = \lim_{n \to \infty}(x_{n+1}) &= \lim_{n \to \infty} \left( x_n - \frac{ f(x_n) }{ f'(x_n) } \right)  = \xi - \frac{ f(\xi) }{ f'(\xi) }. \end{aligned}

Since f'(\xi) \neq 0, we must have f(\xi) = 0. Therefore, \xi = c, and x_n \to c, as required.

Denote \epsilon_n := |x_n - c|. Further suppose that f'' is continuous.

Problem 5. Determine a universal constant M > 0 such that

\epsilon_{n+1} \leq M \cdot \epsilon_n^2.

(Click for Solution)

Solution. For each n, use Taylor’s theorem to furnish \xi_n \in (c, x_n) such that

\displaystyle 0 = f(c) = f(x_n) + f'(x_n) \cdot (c-x_n) + \frac{f''( \xi_n )}{2} \cdot (c - x_n)^2.

By algebruh,

\displaystyle x_{n+1} - c = -\frac{f(x_n)}{ f'(x_n) } - (c-x_n) = \frac{f''(\xi)}{2f'(x_n)} \cdot (c-x_n)^2.

Taking absolute values,

\displaystyle \epsilon_{n+1} = |x_{n+1} - c| = \frac{f''(\xi)}{2f'(x_n)} \cdot |c-x_n|^2 = \frac{f''(\xi)}{2f'(x_n)} \cdot \epsilon_n^2.

Using the continuity of f' and f'', define

\displaystyle M := \frac 12 \cdot \left( \max_{x \in [0,1]} f''(x) \right) \cdot \left( \max_{x \in [0,1]} \frac 1{f'(x)} \right).

Then \epsilon_{n+1} \leq M \cdot \epsilon_n^2, as required.

—Joel Kindiak, 18 Jan 26, 2026H

,

Published by


Leave a comment