Taylor Series and Remainder Estimates

Approximating complex functions with simple polynomials is one of the most powerful ideas in calculus, with profound implications across physics, engineering, and data science. The Taylor series provides the systematic framework for this approximation, while remainder estimates give us the critical ability to quantify—and control—the error we make when we truncate the infinite series. Mastering these tools allows you to confidently replace intractable functions with polynomials, knowing exactly how accurate your approximation is.

From Local Information to Global Approximation

The core idea of a Taylor series is to represent a smooth function $f(x)$ as an infinite sum of polynomial terms, built from the function's derivatives at a single point. If a function is infinitely differentiable at a point $x=a$, we can construct its Taylor series expansion about $x=a$:

$$f(x)=\sum _{n=0}^{\infty }\frac{f^{(n)}(a)}{n!}(x-a{)}^n.$$

Here, $f^{(n)}(a)$ denotes the $n$-th derivative of $f$ evaluated at $a$, and $n!$ is the factorial of $n$. The special case where $a=0$ is called a Maclaurin series: $f(x)={\sum }_{n=0}^{\infty }\frac{f^{(n)}(0)}{n!}{x}^n$.

Consider a simple example: the exponential function $f(x)=e^x$. All its derivatives are $e^x$, and $e^0=1$. Therefore, its Maclaurin series is beautifully simple: $$e^x=1+x+\frac{x^2}{2!}+\frac{x^3}{3!}+\cdots =\sum _{n=0}^{\infty }\frac{x^n}{n!}.$$

This representation is more than just a formal sum; it is a functional identity that holds for all real numbers $x$. The process of constructing a series is straightforward: compute derivatives, evaluate at the center $a$, and assemble the coefficients.

When Does the Series Actually Represent the Function?

A crucial, often overlooked distinction is between a function being smooth (infinitely differentiable) and being analytic (representable by its convergent Taylor series). A function is analytic at a point $a$ if its Taylor series converges to the function value $f(x)$ for all $x$ in some open interval around $a$. For many common functions like $e^x$, $\sin x$, and $\cos x$, the Taylor series converges to the function everywhere, making them analytic on the entire real line.

However, smoothness does not guarantee analyticity. The classic counterexample is the function: $$f(x)=\{\begin{array}{ll}{e}^{-1/x^2}& \text{if }x\ne 0\\ 0& \text{if }x=0\end{array}.$$ This function is infinitely differentiable at $x=0$, and all its derivatives there are zero. Its Maclaurin series is therefore ${\sum }_{n=0}^{\infty }0\cdot x^n=0$, which converges everywhere—but only equals the original function at the single point $x=0$. This function is smooth but not analytic at $0$.

Thus, the convergence of a Taylor series to the original function is not automatic. We need a tool to bridge the gap between the finite Taylor polynomial $T_n(x)$ (the sum of the first $n+1$ terms) and the full function $f(x)$. This bridge is the remainder or error term, $R_n(x)$.

Quantifying the Error: The Lagrange and Cauchy Remainder Forms

The Taylor polynomial of degree $n$ for $f$ about $a$ is $T_n(x)={\sum }_{k=0}^n\frac{f^{(k)}(a)}{k!}(x-a{)}^k$. The exact identity is $f(x)=T_n(x)+R_n(x)$, where $R_n(x)$ is the remainder. To make the series useful, we must be able to bound $|R_n(x)|$.

The most common form is the Lagrange form of the remainder. It states that if $f$ is $(n+1)$ times differentiable on an interval containing $a$ and $x$, then there exists some number $c$ between $a$ and $x$ such that: $$R_n(x)=\frac{f^{(n+1)}(c)}{(n+1)!}(x-a{)}^{n+1}.$$

This looks like the next term in the series, but with the derivative evaluated at an unknown point $c$. Its power lies in estimation: if you can bound the $(n+1)$-th derivative, you can bound the error. For instance, to approximate $e^{0.1}$ using a 3rd-degree Maclaurin polynomial for $e^x$, we have $T_3(0.1)=1+0.1+(0.1{)}^2/2!+(0.1{)}^3/3!\approx 1.1051667$. The Lagrange remainder is $R_3(0.1)=\frac{e^c}{4!}(0.1{)}^4$, where $0<c<0.1$. Since $e^c<e^{0.1}<1.106$, we get $|R_3(0.1)|<\frac{1.106}{24}\cdot 0.0001\approx 0.00000461$. This confirms our approximation is accurate to at least 5 decimal places.

A related but sometimes sharper form is the Cauchy form of the remainder: $$R_n(x)=\frac{f^{(n+1)}(c)}{n!}(x-c{)}^n(x-a),$$ where, again, $c$ is between $a$ and $x$. This form is particularly useful when the factor $(x-c{)}^n$ allows for a tighter bound than the $(x-a{)}^{n+1}$ factor in the Lagrange form, especially for specific values of $x$ and $a$.

The Integral Form and Practical Estimation Strategies

For functions whose derivatives are integrable, the most precise expression for the remainder is the integral form: $$R_n(x)=\frac{1}{n!}{\int }_a^x(x-t{)}^n{f}^{(n+1)}(t)dt.$$ This is not an estimate but an exact formula. It is derived by repeatedly integrating the derivative of $f$. It's invaluable for deriving other remainder forms and for performing precise analytical work. Applying the Integral Mean Value Theorem to this integral leads directly to the Cauchy form, and a further manipulation yields the Lagrange form.

In practice, your choice of remainder form depends on the function and what you know about its derivatives.

1. Use Lagrange if you have a simple global bound on $|f^{(n+1)}|$ (e.g., $|{\sin }^{(n+1)}(c)|\le 1$).
2. Use Cauchy if the $(x-c{)}^n$ factor offers a significant advantage, such as when $x$ and $a$ are close.
3. Use the Integral Form as a starting point for theoretical proofs or when derivatives are easily integrable.

The ultimate goal is to determine, for a given tolerance $ϵ$ and interval $I$, a polynomial degree $n$ such that $|R_n(x)|<ϵ$ for all $x$ in $I$. This is the essence of remainder estimation.

Common Pitfalls

Assuming Smoothness Implies Analyticity: As shown with the $e^{-1/x^2}$ function, infinite differentiability at a point does not mean the Taylor series converges to the function near that point. Always check the behavior of the remainder $R_n(x)$ as $n\to \infty$.

Misapplying the Lagrange Remainder Bound: The constant $c$ in the Lagrange form is unknown and depends on both $x$ and $n$. A common error is to treat it as a constant when finding a maximum error bound over an interval. You must find the maximum possible value of $|f^{(n+1)}(c)|$ for $c$ in the interval between $a$ and $x$ (or the entire interval of approximation). For $f(x)=\sin x$, the maximum is always 1, but for other functions, it requires careful analysis.

Confusing Convergence of the Series with Convergence to $f(x)$: A Taylor series can converge (i.e., the partial sums have a limit) without converging to the function that generated it. The series for $e^{-1/x^2}$ at $x=0$ converges to 0, but the function is not 0. The remainder term $R_n(x)$ must converge to zero for the series to converge to $f(x)$.

Forgetting the Domain of Analyticity: Even for analytic functions, the Taylor series only converges to the function within its radius of convergence. For example, the Maclaurin series for $\ln (1+x)$, which is ${\sum }_{n=1}^{\infty }\frac{(-1{)}^{n+1}}{n}{x}^n$, converges to the function only for $-1<x\le 1$. Using it outside this interval yields an incorrect result.

Summary

• A Taylor series expands a function into an infinite polynomial based on its derivatives at a point $a$, with the Maclaurin series being the special case where $a=0$.
• A function is analytic at a point only if its Taylor series converges to the function in a neighborhood around that point, a stronger condition than mere smoothness (infinite differentiability).
• The error of a finite Taylor polynomial is captured by the remainder term $R_n(x)$. The Lagrange form $R_n(x)=\frac{f^{(n+1)}(c)}{(n+1)!}(x-a{)}^{n+1}$ is a practical tool for bounding this error if you can bound the size of the derivative.
• The Cauchy form and the exact integral form of the remainder provide alternative, sometimes more useful, expressions for estimating or calculating the error precisely.
• Effective approximation requires using these remainder estimates to select a polynomial degree that guarantees your desired accuracy over the entire interval of interest.