Calculus › Section J · Infinite series › Topic J.5

Alternating series

The harmonic series diverges, yet flip every second sign and it converges. Alternating series show that cancellation is a genuine source of convergence, and they come with the best error estimate in the subject.

Cancellation changes everything

The harmonic series \(\sum \frac{1}{n}\) diverges because its positive terms pile up without bound. Now alternate the signs:

1 - \frac{1}{2} + \frac{1}{3} - \frac{1}{4} + \frac{1}{5} - \cdots

Each new term pushes the running total the other way, so instead of marching upward forever, the total swings back and forth across a target it closes in on. The partial sums of the harmonic series were a one-way staircase; these are a corridor that narrows. A series whose signs alternate, like this one, can converge even when the series of its term sizes diverges. The question is when, and the answer is beautifully simple.

The alternating series test

Write an alternating series as \(\sum (-1)^{n+1} b_n\) with every \(b_n > 0\), so the actual terms are \(b_1, -b_2, b_3, -b_4, \dots\). The alternating series test says the series converges whenever two things hold: the sizes shrink toward zero, \(b_{n+1} \le b_n\) for all \(n\) beyond some point, and they reach it, \(\lim_{n \to \infty} b_n = 0\). Both conditions are needed. If the sizes fail to decrease, the swings can stay wide and miss each other. If they fail to approach 0, the terms jostle the total by undiminished amounts and no settling is possible, as with \(\sum (-1)^{n+1} \frac{n}{n+1}\), whose terms alternate between values near \(+1\) and \(-1\) and whose partial sums jump by nearly 2 forever.

Why it works: a narrowing corridor

Watch the partial sums of \(1 - \frac{1}{2} + \frac{1}{3} - \cdots\): \(S_1 = 1\), \(S_2 = 0.5\), \(S_3 \approx 0.833\), \(S_4 \approx 0.583\), \(S_5 \approx 0.783\), \(S_6 \approx 0.617\). Each odd-indexed sum is larger than the last one, because the new positive term outweighs the negative term added after it; each even-indexed sum is smaller than the last one for the same reason. And every even sum stays below every odd sum, because between them the corridor has narrowed by the two amounts just added. So the odd sums form a decreasing sequence bounded below by every even sum, and the even sums an increasing sequence bounded above by every odd sum, and the gap between the two families is exactly the next term \(b_n\), which shrinks to 0. Both sequences converge, to the same limit: the squeeze from the sequences topic doing its work again.

Worked example: the alternating harmonic series

The series \(\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n}\) passes the test: the sizes \(\frac{1}{n}\) are positive, decreasing and shrinking to 0. So it converges, even though \(\sum \frac{1}{n}\) diverges. Its sum is a number you already know: \(\ln 2 \approx 0.6931\), a fact worth memorising. Check the corridor around that value: \(S_6 \approx 0.617\) lies below it and \(S_5 \approx 0.783\) above it, \(S_8 \approx 0.635\) below and \(S_7 \approx 0.760\) above, and each pair brackets \(\ln 2\) more tightly than the last. The alternating harmonic series converges; it is the standard example of a series rescued purely by cancellation.

The alternating series error bound

The corridor picture hands you an estimate most tests cannot match. If a series satisfies the alternating series test, then the sum \(S\) always lies between two consecutive partial sums, so the error of a partial sum is at most the size of the first omitted term:

|S - S_N| \le b_{N+1}

This is remarkably practical: to hit the sum within \(0.1\), stop at the first place where the next term is at most \(0.1\). For the alternating harmonic series, the term \(\frac{1}{n}\) drops to \(0.1\) at \(n = 10\), so stopping after nine terms, with \(S_9 \approx 0.746\), is guaranteed within \(0.1\) of \(\ln 2\). Indeed \(0.746 - 0.693 = 0.052\), inside the promise. No other test so far turns convergence directly into a guaranteed accuracy with one glance at the next term.

Absolute versus conditional convergence

Now take the absolute values of every term and ask whether that series converges. If \(\sum |a_n|\) converges, the original \(\sum a_n\) converges absolutely, and absolute convergence is the strong kind: it forces the original series to converge too, since cancellation can only help a series whose total magnitude is already finite. If \(\sum a_n\) converges but \(\sum |a_n|\) does not, the series converges conditionally: it converges only because its signs cooperate. The alternating harmonic series is the defining example: it converges, but stripping its signs gives back the divergent harmonic series, so its convergence is conditional. By contrast \(\sum \frac{(-1)^{n+1}}{n^2}\) converges absolutely, because \(\sum \frac{1}{n^2}\) converges on its own.

Why the distinction matters: rearrangements

The difference between the two kinds of convergence is not bookkeeping; it changes what you are allowed to do. An absolutely convergent series is sturdy: its terms can be rearranged in any order, grouped however you like, and the sum never changes. A conditionally convergent series is eerie: Riemann proved its terms can be rearranged to produce any target number you choose, or to diverge. The mechanism is that the positive terms alone diverge and the negative terms alone diverge, so a rearranger can spend more positive terms early to climb toward a high target, then pay with negatives to level off, indefinitely. The alternating harmonic series can be shuffled into \(\frac{3}{2}\ln 2\), or \(100\), or anything else. Order matters, so for conditionally convergent series you keep the order you were given.

Worked example: classifying by kind of convergence

Run the classification on examples. \(\sum \frac{(-1)^{n+1}}{n^2}\): strip signs to get the convergent p-series \(\sum \frac{1}{n^2}\), so it converges absolutely. \(\sum \frac{(-1)^{n+1}}{\sqrt{n}}\): sizes \(\frac{1}{\sqrt{n}}\) decrease to 0, so the alternating test gives convergence, but \(\sum \frac{1}{\sqrt{n}}\) diverges (\(p = \frac{1}{2}\)), so only conditional. \(\sum \frac{(-1)^{n+1}}{2^n}\): stripping signs leaves a convergent geometric series, so absolute. The classification always answers two questions in order: does it converge at all (divergence test, alternating test, or a comparison on the absolute values), and then does it converge absolutely? If the first answer is no, the second is moot.

Common mistakes to avoid

The recurring error is checking that the terms shrink to 0 while forgetting to check that their sizes decrease; the test needs both, and a series whose sizes wobble can diverge even with terms approaching 0. Second, never apply the error bound to a series the alternating test has not verified: the corridor argument is what makes \(|S - S_N| \le b_{N+1}\) true. Third, do not say a series "converges by the alternating test" when the signs do not actually alternate: the test is for series of the form \(\sum (-1)^n b_n\) or \(\sum (-1)^{n+1} b_n\) with \(b_n > 0\). Fourth, keep the vocabulary straight: absolute means the sign-stripped series converges; conditional means convergence depends on the signs; neither applies to a divergent series.

Key takeaways

An alternating series \(\sum (-1)^{n+1} b_n\) with positive, decreasing sizes shrinking to 0 converges, because its odd partial sums descend and its even partial sums climb into a narrowing corridor with one shared limit. Its sum always lies between consecutive partial sums, so the error after \(N\) terms is at most the first omitted term, \(b_{N+1}\). Absolute convergence means \(\sum |a_n|\) converges and makes the original order-proof; conditional convergence means the signs are doing essential work, and Riemann's rearrangement fact shows such series must keep their given order. The alternating harmonic series converges to \(\ln 2\) conditionally.

Flashcards

Flashcard

Alternating series

Signs alternate: Σ (-1)ⁿ⁺¹ bₙ with bₙ > 0

Term sizes carry the information.

Test your understanding

Question 1 of 5

What does the series 1 - 1/2 + 1/3 - 1/4 + … converge to?

Question 2 of 5

What can be said about the partial sum S₉ of the alternating harmonic series?

Question 3 of 5

Does the series Σ (-1)ⁿ⁺¹ (n/(n+1)) converge?

Question 4 of 5

The series Σ (-1)ⁿ⁺¹/n³ converges. What kind of convergence is it?

Question 5 of 5

Which series converges conditionally?

Standards: 10 (BC)

← Previous topicNext topic →