The alternating series test is one of the cleanest convergence tools in the AP Calculus syllabus, and it travels unusually well between the IB Diploma programme and AP-style sequences-and-series questions. Most IB Math AA HL candidates meet the test first inside a sub-topic of Paper 2 sequences and series, then encounter it again on AP Calculus when the topic returns as a free-response item. The version IB students see tends to emphasise conceptual reasoning: does the sequence decrease, does the term tend to zero, and what does convergence actually buy you. The AP version sharpens that picture with a free-response requirement to estimate the remainder, justify truncation choices, and connect the test to alternating-error bounds. Getting the alternating series test right is one of the higher-leverage moves a student can make, because a single correct application can replace dozens of mechanical lines elsewhere on the paper.
The statement of the alternating series test: what is actually being claimed
Most candidates who lose marks on this topic have memorised the statement but not the structure. The alternating series test, sometimes called the Leibniz test, applies to infinite series of the form ∑(−1)^n a_n, where a_n ≥ 0 for every n in the relevant domain. There are two conditions, and both must be satisfied for the test to confirm convergence. The first is that the sequence a_n must be monotonically decreasing once the series is in a position to be tested — usually after a finite number of terms. The second is that the limit of a_n as n approaches infinity must equal zero. Either condition on its own is not enough. A series whose terms do not tend to zero cannot converge, and a series whose terms tend to zero but do not decrease can still diverge, so the test deliberately demands both.
For IB students, the most useful restatement is a logical one. The test gives a sufficient condition for convergence, not a necessary one. If both conditions hold, the series converges. If a condition fails, the test gives no information and the candidate must move to a different tool, such as the comparison test, the ratio test, or the integral test. The test is silent on the sum of the series: knowing a series converges tells the candidate nothing about its value. That is why AP free-response questions pair the test with a remainder estimate. A candidate who writes "by the alternating series test the series converges, therefore the sum is approximately 1.5" has conflated convergence with evaluation, which is a recurring mark-loss in marking schemes.
One more piece of the statement is worth memorising carefully. The test requires a_n to be decreasing, not strictly decreasing. A constant tail — a_n = 0.1 for all n beyond some index — still satisfies the test if it stays non-negative and tends to zero. Candidates who insist on strict inequality sometimes write "the sequence is not strictly decreasing, so the test fails," which costs marks on questions where the test clearly does apply. The decreasing condition exists because the alternating signs need a monotonically shrinking positive term to ensure the partial sums oscillate inside a narrowing window. The next paragraph makes that mechanism visible.
Mentally picture the partial sums. If a_1, a_2, a_3, … are positive and decreasing to zero, then S_1 = a_1 sits above the eventual sum, S_2 = a_1 − a_2 drops below, S_3 climbs back up but not as high as S_1, and so on. The even partial sums form an increasing sequence bounded above, and the odd partial sums form a decreasing sequence bounded below, and both squeeze toward the same limit. The limit exists precisely because the decreasing condition holds. Without monotonicity, the partial sums can wobble outside any shrinking window and the limit need not exist. This geometric picture is what examiners want to see hinted at when the question is worth a high mark. A line such as "the partial sums oscillate within a window of width a_{N+1}, which tends to zero" carries more weight than a flat restatement of the theorem.
Worked example 1: ∑(−1)^n / (n + 1) from n = 1 to infinity
The series ∑(−1)^n / (n + 1) is a friendly starting case and almost identical to the IB textbook's lead example. The terms alternate in sign because of the (–1)^n factor, and the absolute terms are a_n = 1 / (n + 1). The candidate must check two things. First, is a_n decreasing? Yes: the denominator grows by 1 with each step, so 1/(n + 1) gets smaller monotonically. Second, does a_n tend to zero? Yes, because 1/(n + 1) → 0 as n → ∞. Both conditions hold, so the alternating series test confirms convergence. The conclusion is not that the series converges absolutely — in fact the corresponding positive series ∑ 1/(n + 1) is the harmonic series shifted by an index, which diverges. Convergence here is conditional.
A strong AP-style justification expands the test. The candidate should write, in order: identify the form, name the positive term a_n, verify monotonicity explicitly (for instance by showing a_{n+1} − a_n = −1/[(n + 1)(n + 2)] < 0), confirm that the limit of a_n is zero (typically by citing that 1/(n + 1) is the reciprocal of an unbounded sequence), and then state the conclusion. Marking schemes reward each line, not just the conclusion. IB Paper 2 marks for sequences and series often follow a similar rubric, awarding 1 mark for the form, 1 mark for monotonicity, 1 mark for the limit, and 1 mark for the conclusion.
The same series, viewed through a free-response lens, opens up a follow-up: estimate the sum using the first four non-zero terms and bound the error. S_4 = 1/2 − 1/3 + 1/4 − 1/5 = 0.2833…, and the alternating series remainder theorem guarantees that the error |S − S_4| is bounded above by the first omitted term, 1/6 ≈ 0.1667. So the true sum lives in [0.1167, 0.4500]. Notice that this bound is not tight — the actual sum is ln(2) − 1 ≈ 0.6931 − 1, but that equals −0.3069, which is outside the bound. The reason is that the partial sums of this shifted harmonic start at n = 1, so the index for the remainder bound needs careful treatment. IB candidates moving to AP should expect such index traps. The cleaner starting index n = 0 gives ∑(−1)^n / (n + 1) = ln(2) ≈ 0.6931, which lies inside any reasonable bound, and is the form most textbooks prefer.
Worked example 2: ∑(−1)^n · n / (n^2 + 1) — the trap that catches most students
This second example is the one I would put in front of any IB candidate transitioning to AP-style questions, because it shows exactly where the alternating series test is not the right tool. The terms alternate, and a_n = n / (n^2 + 1) is positive. A naive read says: try the alternating series test. The candidate checks a_n → 0 (true: divide top and bottom by n^2 to get 1/(n + 1/n), which tends to zero). The candidate checks monotonicity by computing a_{n+1} − a_n or by considering a_n = 1/(n + 1/n) and noting the denominator is increasing. The test appears to apply. So the candidate writes "convergent by the alternating series test" and moves on. They have just lost a mark, because the test was the wrong tool and the conclusion is correct for the wrong reason.
The series is not just convergent; it is absolutely convergent. The corresponding positive series ∑ n / (n^2 + 1) behaves like ∑ 1/n for large n, which is the harmonic series and diverges — so absolute convergence fails by direct comparison. But absolute divergence does not contradict the alternating series test. Both conclusions are consistent. The test only licenses a conclusion of convergence, and the candidate should have stopped at "converges conditionally." Where IB students go wrong is the next step. AP free-response items often ask, in part (b), whether the series converges absolutely. If a candidate has glued the alternating series test to absolute convergence, the second part unravels. The fix is to separate two questions in writing: does the alternating series test apply? and does the corresponding positive series converge?. The answers are independent.
The cleanest answer to part (b) on this example is to invoke the limit comparison test against the harmonic series. Compute lim_{n→∞} [n/(n^2 + 1)] / [1/n] = lim_{n→∞} n^2 / (n^2 + 1) = 1, a finite non-zero limit, so both series share the same convergence behaviour. Since the harmonic series diverges, the positive series ∑ n/(n^2 + 1) diverges, and the alternating series converges conditionally but not absolutely. The candidate who writes this clearly has shown command of three convergence tools in two lines, which is exactly what the AP rubric rewards. For most candidates I coach, this is the first place where the alternating series test stops being an answer and starts being a single step in a longer argument.
Two competing tools: where the ratio test and the integral test take over
Once a candidate has internalised that the alternating series test only licenses a conditional convergence conclusion, the natural next question is: which test should I reach for first? A useful rule of thumb, and one I share in tutoring sessions, is to look at the form of a_n. If a_n is a ratio of polynomials, factorials, or exponentials, the ratio test is almost always cleaner than the alternating series test. If a_n is something like 1/(n ln n) or arctan(n)/n^2, the integral test often wins. The alternating series test is best reserved for the moment when the sign pattern is the only thing keeping the series alive — that is, when the positive series diverges and only the alternation rescues convergence.
Consider ∑(−1)^n · n! / n^n. The ratio test on the positive series gives lim |a_{n+1}/a_n| = lim [(n+1)!/(n+1)^{n+1}] · [n^n/n!] = lim n^n / (n+1)^n = lim 1/(1 + 1/n)^n = 1/e < 1, so the positive series converges absolutely. The alternating series test would also confirm convergence, but the conclusion would be weaker. The candidate who reaches for the alternating series test first has left a mark on the table: the question was about absolute convergence, and the ratio test answers that directly. The lesson: the alternating series test is the right tool when the positive part is divergent and the test is what saves the day, not when the positive part is already convergent on its own.
The integral test is a different story. For a series like ∑(−1)^n / (n^2 + n + 1), the alternating series test confirms convergence, and the integral test on the positive part ∫ dx / (x^2 + x + 1) converges (it is a bounded integral over [1, ∞)), which would imply absolute convergence. But computing the integral is wasteful compared to the direct comparison: 1/(n^2 + n + 1) < 1/n^2 and ∑ 1/n^2 converges. So the integral test here is overkill. The pattern candidates should internalise: the integral test is most useful when the function is a clean logarithm, square root, or arctangent that integrates easily, and the comparison or limit comparison is a fallback when the integral is messy.
Comparison table: which convergence test to reach for first
| Form of a_n | Best first test | Why | What the alternating series test adds |
|---|---|---|---|
| Polynomial ratio, e.g. n / (n^2 + 1) | Limit comparison with 1/n or 1/n^2 | Same growth order is obvious | Confirms conditional convergence if the positive part diverges |
| Factorial or exponential, e.g. n! / n^n | Ratio test | Factorials and exponentials telescope cleanly | Unnecessary unless signs become the only thing keeping the sum finite |
| Logarithm or arctangent, e.g. 1/(n ln n) | Integral test or direct comparison | Integrals of 1/x or arctan(x) are standard | Useful when the positive part diverges, e.g. ∑(−1)^n / (n ln n) |
| Trigonometric, e.g. sin(n)/n^2 | Absolute value comparison with 1/n^2 | |sin(n)| ≤ 1 is the key bound | Often skipped, since absolute convergence follows immediately |
| Alternating with non-zero limit, e.g. (–1)^n · (1 + 1/n) | nth-term test (divergence) | Terms do not tend to zero | Does not apply; use the nth-term test instead |
Common pitfalls and how to avoid them
Five recurring errors surface year after year, both on IB Paper 2 and on AP-style free-response. The first is treating the alternating series test as if it were a stand-alone sufficient condition for absolute convergence. It is not. A candidate who writes "by the alternating series test, the series converges absolutely" has confused conditional and absolute convergence. The test is silent on absolute convergence; the candidate must test the positive series separately, usually with the comparison, limit comparison, or ratio test. The fix is mechanical: write the conclusion as "the series converges conditionally by the alternating series test," and add a separate line for absolute convergence.
The second pitfall is forgetting the limit-of-terms condition. A series like ∑(−1)^n · (1 + 1/n) has alternating signs and a monotonically decreasing a_n, but a_n → 1 ≠ 0, so the test does not apply. The candidate who applies the test anyway reaches a false conclusion. The fix is to compute lim a_n before invoking the test, even on questions where the limit is obviously zero. The two-line check takes five seconds and prevents a major mark loss. The nth-term test for divergence is the natural fallback here: if a_n does not tend to zero, the series diverges, period.
The third pitfall is misreading the sign pattern. A series like ∑(−1)^{n+1} · 1/n starts positive, then alternates. A series like ∑(−1)^n · 1/n starts negative, then alternates. The test does not care which sign comes first, but the candidate's partial sums do. When a free-response question asks for an approximation to the sum, the candidate must use the first term whose sign matches the requested parity. A candidate who approximates ∑(−1)^n · 1/n with the first three terms gets 1 − 1/2 + 1/3 = 0.8333, when the actual partial sum of the first three terms is −1 + 1/2 − 1/3 = −0.8333. The sign pattern matters. The fix is to write the first three or four terms explicitly before computing the partial sum.
The fourth pitfall is monotonicity at the boundary. The test requires a_n to be decreasing eventually, not necessarily from the first term. A series like ∑(−1)^n · (n − 5)/(n + 5) has a_n negative for the first ten terms, which means a_n is not positive, so the test cannot be applied in the standard form. The candidate must either rearrange to start at n = 11, where a_n becomes positive, or use a different test. The fix is to check the sign of a_n for the index range given in the question. Most marking schemes explicitly test for this; missing the sign check is one of the highest-frequency errors IB students make on this topic.