Calculus II: Taylor Polynomial Error Bounds

A Taylor polynomial approximates a function, but an approximation without a known error bound is often useless in engineering. Whether you're modeling stress in a beam, predicting orbital trajectories, or designing a control system, you must know how accurate your polynomial model is. The two primary tools—the Lagrange remainder and the Alternating Series Estimation Theorem—allow you to quantify this error and make reliable, confident computations.

Understanding the Lagrange Remainder Formula

When you approximate a function $f(x)$ with its $n$th-degree Taylor polynomial $T_n(x)$ centered at $x=a$, the exact value is $f(x) = T_n(x) + R_n(x)$. The term $R_n(x)$ is the remainder or error. The Lagrange remainder formula provides a powerful way to bound this error. It states that if $f$ is differentiable $n+1$ times on an interval containing $a$ and $x$, then the error is given by: $$R_n(x) = \frac{f^{(n+1)}(c)}{(n+1)!}(x-a)^{n+1}$$ for some number $c$ between $a$ and $x$.

You don't need to find the exact $c$; you only need to find its maximum possible influence on the error. This is a crucial shift in thinking: instead of finding the exact error, you find a worst-case bound. The formula tells you the error looks like the next term in the Taylor series, but with the $(n+1)$th derivative evaluated at a mysterious point $c$ instead of at the center $a$.

Finding Maximum Error Over an Interval

For practical applications, you usually need the maximum error over an entire range of $x$-values, not just at a single point. This involves a three-step optimization process:

1. Identify the Interval: Determine the interval $I$ over which you are approximating the function. For example, you might need to approximate $\sin(x)$ on $[0, 0.5]$ radians.
2. Bound the Derivative: Find a number $M$ such that $|f^{(n+1)}(c)| \le M$ for all possible $c$ in the interval $I$. This $M$ is the maximum absolute value of the $(n+1)$th derivative on that interval.
3. Compute the Bound: Plug $M$ into the Lagrange formula to get the error bound:

$$|R_n(x)| \le \frac{M}{(n+1)!}|x-a|^{n+1} \quad \text{for all } x \text{ in } I.$$

Worked Example: Approximate $f(x) = e^x$ by $T_2(x)$ (degree 2) centered at $a=0$ on the interval $[-0.5, 0.5]$. We have $T_2(x) = 1 + x + x^2/2$. The remainder uses the third derivative: $f'''(x)=e^x$. On the interval $[-0.5, 0.5]$, the maximum of $|e^c|$ occurs at $c=0.5$, so $M = e^{0.5} \approx 1.6487$. The maximum value of $|x-0|^{3}$ on the interval is $(0.5)^3 = 0.125$. Therefore, the error bound is: $$|R_2(x)| \le \frac{1.6487}{3!} \cdot 0.125 \approx \frac{1.6487}{6} \cdot 0.125 \approx 0.03435.$$ This guarantees our quadratic approximation is within about 0.034 of the true value for any $x$ in $[-0.5, 0.5]$.

Determining the Degree Needed for Desired Accuracy

Often, the engineering requirement is reversed: you have a required accuracy (e.g., error < 0.0001), and you need to find the smallest polynomial degree $n$ that achieves it. This is an iterative process of trial and error with the Lagrange bound.

You set up the inequality $|R_n(x)| < \text{(tolerance)}$ and solve for $n$. Since $n$ is an integer in the factorial, you typically guess values for $n$ until the calculated bound falls below your tolerance. The key is to use the worst-case $x$ and the maximum derivative value $M$ over your interval in the inequality: $$\frac{M}{(n+1)!}|x_{\text{max}}-a|^{n+1} < \text{Tolerance}.$$

The Alternating Series Estimation Theorem

When the Taylor series for a function is an alternating series (terms alternate in sign, decrease in absolute value, and approach zero), a much simpler error bound applies. This is common for series like those for $\sin(x)$, $\cos(x)$, and $e^{-x}$ when $x>0$.

The Alternating Series Estimation Theorem states: If you approximate the sum of a convergent alternating series by its $n$th partial sum, the error is less than or equal to the absolute value of the first omitted term. In Taylor series language, if $T_n(x)$ is the $n$th-degree polynomial and the series is alternating, then: $$|R_n(x)| \le |a_{n+1}|,$$ where $a_{n+1}$ is the $(n+1)$th term of the series.

Example: The series for $\sin(0.1)$ is $0.1 - \frac{(0.1)^3}{3!} + \frac{(0.1)^5}{5!} - ...$. This is alternating. If you use the cubic polynomial $T_3(0.1) = 0.1 - \frac{0.001}{6}$, the error is bounded by the absolute value of the next term: $\frac{(0.1)^5}{120} \approx 8.33 \times 10^{-8}$. This theorem provides an instant, often very tight, error bound without needing to find derivatives.

Practical Error Estimation for Engineering Computations

In practice, engineers blend these tools. The workflow is:

1. Recognize the Series Type: Check if the Taylor series is alternating on your interval. If yes, use the Alternating Series Estimation Theorem—it's simpler and often sharper.
2. Apply Lagrange for General Cases: For non-alternating series (like $e^x$ for positive $x$) or when you need an interval-wide guarantee, use the Lagrange remainder. Carefully find the maximum derivative bound $M$.
3. Validate Computational Models: Before implementing a polynomial approximation in code for a simulation, calculate its error bound over the entire expected input range. This step validates the model's fidelity.
4. Make Trade-off Decisions: Higher-degree polynomials yield smaller errors but require more computation. Error analysis allows you to choose the minimal degree that meets your design specs, optimizing computational efficiency.

Engineering Scenario: You are programming an embedded microcontroller to calculate $\sin(\theta)$ for $\theta$ in $[0, \pi/4]$ using a polynomial. A 5th-degree polynomial gives an error bound of $10^{-5}$ using Lagrange. A 7th-degree polynomial reduces the bound to $10^{-8}$. If your sensor's precision is only $0.001$, the 5th-degree polynomial is more than sufficient and saves precious processor cycles and memory.

Common Pitfalls

1. Misapplying the Alternating Series Theorem: This theorem only works if the series satisfies the three alternating series test conditions (alternating, decreasing absolute terms, limit zero). Applying it to a series like $1 + x + x^2/2! + ...$ for $x>0$ is incorrect because it does not alternate.

• Correction: Always verify the conditions. If the series doesn't alternate, default to the Lagrange remainder.

1. Incorrectly Bounding the Derivative ($M$): The most common mistake with the Lagrange formula is using the derivative at the center $f^{(n+1)}(a)$ instead of finding the maximum $|f^{(n+1)}(c)|$ over the entire interval between $a$ and $x$. Using $a$ can underestimate the true maximum error.

• Correction: Treat $c$ as an unknown in the interval. Use calculus (find critical points) to find the absolute maximum of $|f^{(n+1)}(z)|$ for $z$ in the relevant interval.

1. Confusing the Error Bound with the Actual Error: The Lagrange formula gives an inequality—a ceiling on how bad the error could be. The actual error is often much smaller. Students sometimes think $R_n(x)$ equals the bound, leading to overly pessimistic assessments.

• Correction: Remember the bound is a guarantee: $|\text{Actual Error}| \le \text{Bound}$. It is not a formula for the exact error.

1. Forgetting the Factorial Growth: When solving for the degree $n$ needed for an accuracy, it's easy to underestimate how quickly $(n+1)!$ grows. A small increase in $n$ dramatically reduces the error bound.

• Correction: When iterating to find $n$, increase $n$ incrementally and recalculate; you'll likely find the required $n$ is smaller than a brute-force algebraic approach might suggest.

Summary

• The Lagrange remainder formula, $R_n(x) = \frac{f^{(n+1)}(c)}{(n+1)!}(x-a)^{n+1}$, provides a general, worst-case bound for Taylor polynomial error by finding the maximum value of the relevant derivative over the interval.
• To find a maximum error over an interval, you must bound the $(n+1)$th derivative by a number $M$ across that entire interval before applying the Lagrange formula.
• Determining the polynomial degree for a desired accuracy is an iterative process of applying the Lagrange error bound inequality with increasing values of $n$ until the bound falls below the tolerance.
• The Alternating Series Estimation Theorem offers a simpler, often very tight, error bound ($|R_n| \le |\text{next term}|$) but applies only to Taylor series that satisfy the conditions of the alternating series test.
• In practical engineering computations, error bounding is a critical step for validating polynomial approximations, ensuring they meet design specifications, and making informed trade-offs between accuracy and computational cost.