Taylor's Theorem and Error Bounds

💡 The Core Insight

Taylor's Theorem states that any sufficiently smooth function equals a polynomial (built from its derivatives at a point) plus a precisely quantifiable error term. The error bounds allow us to control approximation quality and are as important as the polynomial itself.

Definition

Taylor's Theorem and Error Bounds: Let $f: [a,b] \to \mathbb{R}$ be a function that is $(n+1)$ times differentiable on the interval $[a,b]$ , and let $c \in [a,b]$ . The $n$ -th degree Taylor polynomial of $f$ centered at $c$ is:

P_n(x) = \sum_{k=0}^{n} \frac{f^{(k)}(c)}{k!}(x-c)^k

The Taylor remainder (or error) is $R_n(x) = f(x) - P_n(x)$ . Taylor's Theorem provides exact formulas for this remainder, the most common being:

Lagrange form: $R_n(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!}(x-c)^{n+1}$ for some $\xi$ between $c$ and $x$
Cauchy form: $R_n(x) = \frac{f^{(n+1)}(\xi)}{n!}(x-\xi)^n(x-c)$ for some $\xi$ between $c$ and $x$
Integral form: $R_n(x) = \int_c^x \frac{f^{(n+1)}(t)}{n!}(x-t)^n \, dt$

TheoremMain Result

Theorem (Taylor's Theorem with Lagrange Remainder) [1]: Let $f: [a,b] \to \mathbb{R}$ be a function such that: 1. $f$ is $n$ times continuously differentiable on $[a,b]$ 2. $f^{(n+1)}$ exists on $(a,b)$ 3. $c \in [a,b]$ Then for any $x \in [a,b]$ , there exists a point $\xi$ strictly between $c$ and $x$ (or $\xi = c$ if $x = c$ ) such that:

f(x) = \sum_{k=0}^{n} \frac{f^{(k)}(c)}{k!}(x-c)^k + \frac{f^{(n+1)}(\xi)}{(n+1)!}(x-c)^{n+1}

Corollary (Error Bound) [2]: If $|f^{(n+1)}(t)| \leq M$ for all $t$ between $c$ and $x$ , then:

|R_n(x)| = |f(x) - P_n(x)| \leq \frac{M}{(n+1)!}|x-c|^{n+1}

Proof

We prove the theorem using Cauchy's Mean Value Theorem. Define the auxiliary function:

g(t) = f(x) - \sum_{k=0}^{n} \frac{f^{(k)}(t)}{k!}(x-t)^k

This function is constructed so that $g(x) = 0$ and $g(c) = f(x) - P_n(x) = R_n(x)$ . Taking the derivative with respect to $t$ :

g'(t) = -\frac{f^{(n+1)}(t)}{n!}(x-t)^n

This follows from the telescoping cancellation: when we differentiate the sum, each term $\frac{f^{(k)}(t)}{k!}(x-t)^k$ produces a contribution that cancels with part of the next term's derivative. Now define a second auxiliary function:

h(t) = (x-t)^{n+1}

Then $h(x) = 0$ , $h(c) = (x-c)^{n+1}$ , and $h'(t) = -(n+1)(x-t)^n$ . Both $g$ and $h$ are continuous on $[c,x]$ (or $[x,c]$ if $x < c$ ) and differentiable on the interior. By Cauchy's Mean Value Theorem, there exists $\xi$ strictly between $c$ and $x$ such that:

\frac{g(x) - g(c)}{h(x) - h(c)} = \frac{g'(\xi)}{h'(\xi)}

Substituting our known values:

\frac{0 - R_n(x)}{0 - (x-c)^{n+1}} = \frac{-\frac{f^{(n+1)}(\xi)}{n!}(x-\xi)^n}{-(n+1)(x-\xi)^n}

Simplifying:

\frac{R_n(x)}{(x-c)^{n+1}} = \frac{f^{(n+1)}(\xi)}{(n+1)!}

Therefore:

R_n(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!}(x-c)^{n+1}

The error bound corollary follows immediately by taking absolute values and using the bound on $f^{(n+1)}$ . □

∎

⚠️ Common mistake

Common Misconception: The value $\xi$ in the remainder formula is not computable in general—we only know it exists somewhere between $c$ and $x$ . Students often try to "find $\xi$ " but this misses the point: the theorem is useful because it gives us bounds on $|R_n(x)|$ without knowing $\xi$ exactly. We use bounds on $|f^{(n+1)}|$ over the entire interval to control the error.

mathematician

Brook Taylor (1715) — Published "Methodus Incrementorum Directa et Inversa" containing the first general statement of the Taylor series expansion, though without rigorous remainder estimates.

visualization

Example1

Approximating $e^{0.1}$ with error bound Let $f(x) = e^x$ and $c = 0$ . We'll use the second-degree Taylor polynomial to approximate $e^{0.1}$ . First, compute derivatives:

$f(x) = e^x$ , so $f(0) = 1$
$f'(x) = e^x$ , so $f'(0) = 1$
$f''(x) = e^x$ , so $f''(0) = 1$
$f'''(x) = e^x$

The Taylor polynomial is:

P_2(x) = 1 + x + \frac{x^2}{2}

Therefore:

P_2(0.1) = 1 + 0.1 + \frac{(0.1)^2}{2} = 1 + 0.1 + 0.005 = 1.105

For the error bound, we need to bound $|f'''(\xi)|$ for $\xi \in [0, 0.1]$ . Since $f'''(x) = e^x$ is increasing, the maximum occurs at $x = 0.1$ :

|f'''(\xi)| \leq e^{0.1} < e^{1} < 3

Thus:

|R_2(0.1)| \leq \frac{3}{3!}(0.1)^3 = \frac{3}{6}(0.001) = 0.0005

Our approximation: $e^{0.1} \approx 1.105 \pm 0.0005$ , i.e., $1.1045 \leq e^{0.1} \leq 1.1055$ . (The actual value is $e^{0.1} \approx 1.10517$ , well within our bounds!)

Example2

Finding how many terms are needed for a given accuracy Suppose we want to approximate $\sin(1)$ with error less than $10^{-6}$ using a Taylor polynomial centered at $c = 0$ . For $f(x) = \sin(x)$ , the derivatives cycle: $\sin, \cos, -\sin, -\cos, \sin, \ldots$ The Taylor series is:

\sin(x) = x - \frac{x^3}{3!} + \frac{x^5}{5!} - \frac{x^7}{7!} + \cdots

For all $k$ , we have $|f^{(k)}(x)| \leq 1$ . The Lagrange remainder for degree $n$ satisfies:

|R_n(1)| \leq \frac{1}{(n+1)!}

We need $\frac{1}{(n+1)!} < 10^{-6}$ , so $(n+1)! > 10^6$ . Checking: $7! = 5040$ , $8! = 40320$ , $9! = 362880$ , $10! = 3628800 > 10^6$ . Thus $n = 9$ suffices. Using $P_9(x) = x - \frac{x^3}{6} + \frac{x^5}{120} - \frac{x^7}{5040} + \frac{x^9}{362880}$ :

P_9(1) = 1 - \frac{1}{6} + \frac{1}{120} - \frac{1}{5040} + \frac{1}{362880} \approx 0.8414710

This approximates $\sin(1) \approx 0.8414709848$ with the required accuracy.

Counterexample

Why differentiability is essential Consider $f(x) = |x|$ centered at $c = 0$ . We might naively write $P_1(x) = 0 + 0 \cdot x = 0$ since $f(0) = 0$ and we might think " $f'(0) = 0$ ." But $f'(0)$ does not exist—the left and right derivatives differ. The function has a corner at $x = 0$ . There is no first-degree polynomial that approximates $|x|$ well near $0$ with error going to zero faster than $|x|$ itself. Indeed, for any linear function $ax + b$ , we have:

\left||x| - (ax+b)\right| \not= o(|x|) \text{ as } x \to 0

This shows that without differentiability at $c$ , we cannot even construct the first-degree Taylor polynomial meaningfully, let alone guarantee the error bounds. The smoothness hypotheses in Taylor's Theorem are essential.

🧠 Intuition

Geometric Picture: Imagine zooming in on a smooth curve near a point. As you zoom closer and closer, the curve looks more and more like a straight line (the tangent). Zooming in further, you might notice slight curvature—this is captured by the quadratic term. Each additional Taylor term captures the next level of detail in the curve's shape. The remainder term $R_n(x)$ measures "what's left over" after accounting for polynomial behavior up to degree $n$ . The factor $\frac{(x-c)^{n+1}}{(n+1)!}$ explains why the approximation gets better as we take more terms (factorial growth in denominator) and why it's better closer to $c$ (the power $(x-c)^{n+1}$ gets small). Think of $P_n$ as a "polynomial microscope" with $n$ levels of magnification, and $R_n$ as the resolution limit.

connection

Numerical Analysis and Computational Mathematics: Taylor's Theorem is the foundation of most numerical differentiation and integration schemes. For instance, the Euler method for solving differential equations is essentially a first-order Taylor approximation: $y(x+h) \approx y(x) + hy'(x)$ . Higher-order methods like Runge-Kutta methods use more Taylor terms implicitly. The error bounds allow us to prove convergence rates and choose appropriate step sizes. This connection shows that theoretical analysis directly enables practical computation—the $O(h^{n+1})$ error bound translates into concrete performance guarantees for algorithms.

Taylor's Theorem and Error Bounds

Taylor's Theorem and Error Bounds

Overview

Core Theory

Examples and Applications

Historical Development

Connections to Other Areas

What You Need First

What This Opens Up

See Also