Calculus II: Alternating Series

When you need to compute the infinite sum of a series, you’re often faced with a practical problem: where do you stop adding terms to get a usable answer? For series whose terms alternate between positive and negative, calculus provides exceptionally powerful and precise tools for both determining convergence and estimating the error involved in using a partial sum. Mastering alternating series is crucial for engineers, as they appear in error analysis, signal processing, and numerical solutions to differential equations, where understanding the accuracy of an approximation is as important as the calculation itself.

The Alternating Series Test (Leibniz's Test)

An alternating series is one whose terms alternate in sign. The general form is ${\sum }_{n=1}^{\infty }(-1{)}^{n-1}{b}_n$ or ${\sum }_{n=1}^{\infty }(-1{)}^n{b}_n$, where $b_n>0$ for all $n$. The Alternating Series Test (or Leibniz's Test) gives us a straightforward condition for convergence.

*An alternating series $\sum (-1{)}^{n-1}{b}_n$ converges if both of the following two conditions are met:*

1. $b_{n+1}\le b_n$ for all $n$ (The sequence $b_n$ is decreasing).
2. ${\lim }_{n\to \infty }{b}_n=0$.

It is critical to check both conditions. For example, consider the alternating harmonic series: ${\sum }_{n=1}^{\infty }(-1{)}^{n-1}\frac{1}{n}$. Here, $b_n=\frac{1}{n}$. First, $\frac{1}{n+1}<\frac{1}{n}$, so the terms are decreasing. Second, ${\lim }_{n\to \infty }\frac{1}{n}=0$. Both conditions are satisfied, so the series converges. However, the test only tells us that it converges, not what it converges to. For the alternating harmonic series, the sum is $\ln (2)$.

Error Bound for Partial Sums

The true power of the Alternating Series Test for applied work lies in its built-in error estimate. If an alternating series satisfies the conditions of the test and converges to a sum $S$, then the error involved in approximating $S$ by the $n$th partial sum $s_n$ is bounded by the absolute value of the first neglected term.

Alternating Series Estimation Theorem: If an alternating series $\sum (-1{)}^{n-1}{b}_n$ satisfies the conditions for convergence, and its sum is $S$, then the error $|R_n|=|S-s_n|$ satisfies: $$|R_n|\le b_{n+1}.$$

This means if you stop adding terms after the $n$th term, your approximation is guaranteed to be within $b_{n+1}$ of the true sum. This is exceptionally useful for computation. For instance, suppose you are approximating ${\sum }_{n=1}^{\infty }\frac{(-1{)}^{n-1}}{n^3}$ using the first 4 terms ($s_4$). The error bound is $b_5=\frac{1}{125}=0.008$. You know immediately that $s_4$ is within 0.008 of the true sum, without knowing the true sum itself.

Absolute Versus Conditional Convergence

This distinction is fundamental to understanding the stability of a series' sum. A series $\sum a_n$ is absolutely convergent if the series of absolute values $\sum |a_n|$ converges. A series is conditionally convergent if $\sum a_n$ converges but $\sum |a_n|$ diverges.

Every absolutely convergent series is convergent, but the converse is not true. The alternating harmonic series is the classic example of conditional convergence: $\sum (-1{)}^{n-1}/n$ converges, but the harmonic series $\sum 1/n$ diverges. Why does this matter? Absolute convergence is a stronger form of convergence that guarantees the series sum will remain unchanged if the terms are rearranged. Conditional convergence does not offer this guarantee, leading to surprising and counter-intuitive results.

The Riemann Rearrangement Theorem

The Rearrangement Theorem states a profound result: if a series is conditionally convergent, its terms can be rearranged to converge to any real number whatsoever, or even to diverge. This has significant implications for numerical computation.

The intuition is that a conditionally convergent series, like the alternating harmonic series, relies on a precise balance between its positive and negative terms to achieve a finite sum. By changing the order, you can manipulate this balance. For example, you could rearrange terms to sum all the positive terms first, making the partial sums grow very large, then occasionally subtract a negative term to "steer" the sum toward a target number. In practice, this means the commutative property of addition does not hold for conditionally convergent series. For absolutely convergent series, any rearrangement converges to the same sum, making them much more "well-behaved" for mathematical modeling.

Practical Error Estimation for Computing Series Sums

For an engineer, the theory culminates in a practical workflow for computing a series sum to a desired accuracy, $ϵ$.

Step 1: Identify the Series Type. Confirm you have an alternating series of the form $\sum (-1{)}^{n-1}{b}_n$ where $b_n>0$.

Step 2: Verify the Alternating Series Test Conditions. Check that $b_n$ is decreasing (often using derivatives or simple inequalities) and that $b_n\to 0$.

Step 3: Apply the Error Bound. To find the partial sum $s_n$ that approximates the true sum $S$ with an error less than $ϵ$, solve the inequality $b_{n+1}<ϵ$ for $n$.

Step-by-Step Example: Compute the sum of ${\sum }_{n=1}^{\infty }\frac{(-1{)}^{n-1}}{n^4}$ with an error less than 0.0001.

1. This is an alternating series with $b_n=1/n^4$.
2. The terms are positive, decreasing ($1/(n+1{)}^4<1/n^4$), and approach zero.
3. We need $b_{n+1}=\frac{1}{(n+1{)}^4}<0.0001$.

Solving: $(n+1{)}^4>10,000$, so $n+1>10$. Thus, $n>9$. Therefore, using the partial sum $s_{10}$ (the sum of the first 10 terms) guarantees an error less than 0.0001. You would compute $s_{10}$ directly to get your approximation.

This process transforms an infinite calculation into a finite, manageable one with a known precision—a routine necessity in engineering simulations and numerical analysis.

Common Pitfalls

1. Misapplying the Alternating Series Test Conditions. The most frequent error is checking only the limit condition ($\lim b_n=0$) and forgetting to verify that the sequence $b_n$ is decreasing. The series ${\sum }_{n=1}^{\infty }(-1{)}^{n-1}\frac{n}{2n+1}$ fails because while the terms go to $1/2$, not 0, the decreasing condition is also not satisfied. Always check both.
2. Confusing the Error Bound Formula. Remember, the error bound $|R_n|$ is at most the first omitted term, $b_{n+1}$. Students often mistakenly use $b_n$ (the last term included) or think it's an exact error, not just an upper bound. If $b_{n+1}=0.01$, your error is less than or equal to 0.01, not necessarily exactly 0.01.
3. Overlooking Absolute vs. Conditional Convergence. When performing operations on a series (like multiplying it by another series or function), assuming properties that only hold for absolutely convergent series can lead to incorrect results. Always test $\sum |a_n|$ if the stability of the sum under manipulation is important to your application.
4. Incorrectly Applying to Non-Alternating Series. The alternating series error bound is a special property. You cannot use $b_{n+1}$ as an error bound for a series that is not alternating and satisfying the Leibniz conditions. For other convergent series (e.g., a positive-term series converged by the Integral Test), you must use that test's specific remainder estimation technique.

Summary

• The Alternating Series Test (Leibniz's Test) provides simple conditions for convergence: terms must decrease monotonically and approach zero.
• Its greatest practical strength is the built-in error bound: the error of a partial sum approximation is always less than or equal to the first omitted term ($|R_n|\le b_{n+1}$).
• Absolute convergence ($\sum |a_n|$ converges) is a stronger, more stable form of convergence than conditional convergence ($\sum a_n$ converges but $\sum |a_n|$ diverges).
• The Riemann Rearrangement Theorem reveals a critical limitation: the sum of a conditionally convergent series can change if its terms are rearranged, a property not shared by absolutely convergent series.
• The standard workflow for practical computation involves verifying the alternating series conditions, then using the inequality $b_{n+1}<ϵ$ to determine how many terms are needed to achieve a desired accuracy $ϵ$.