Why Hypotheses Matter
Convergence tests are conditional statements: they give a conclusion only when their stated hypotheses hold. A calculation that resembles a test is not enough. Before using a result, check the type of terms involved, the required inequality or limit, and any assumptions about a related function.
One principle applies throughout: changing or removing finitely many terms does not change whether a series converges. Thus, hypotheses that hold from some index onward are sufficient, even if they fail among the first few terms. But a condition that fails repeatedly in the tail cannot be dismissed as an initial exception.
What the Tests Require
The tests in this tutorial have different hypotheses and different conclusions. In particular, inequalities in direct comparison have a required direction, while a limit of one in the Ratio Test does not settle convergence.
| Test | Conditions and conclusion |
|---|---|
| Integral Test | \(a_n=f(n)\), where \(f\) is positive, continuous, and nonincreasing eventually. Then \(\sum a_n\) and \(\int_N^\infty f(x)\,dx\) either both converge or both diverge. |
| Direct Comparison | For eventually nonnegative terms, \(0\le a_n\le b_n\) and convergence of \(\sum b_n\) imply convergence of \(\sum a_n\). If \(0\le b_n\le a_n\) and \(\sum b_n\) diverges, then \(\sum a_n\) diverges. |
| Limit Comparison | For eventually positive terms, if \(\lim_{n\to\infty}a_n/b_n=c\) with \(0<c<\infty\), then \(\sum a_n\) and \(\sum b_n\) have the same convergence behavior. |
| Alternating Series Test | For terms with alternating signs and magnitudes \(b_n\ge0\), eventual nonincrease and \(\lim b_n=0\) imply convergence. |
| Ratio Test | If the limit of \(\lvert a_{n+1}/a_n\rvert\) exists, a value below \(1\) implies absolute convergence; a value above \(1\) implies divergence. A value of \(1\) is inconclusive. |
Integral Test: A Function Must Actually Decrease
The Integral Test does not say that convergence of any integral associated with a sequence determines convergence of the series. The function must be positive, continuous, and nonincreasing on the tail, and its values at the positive integers must be the terms of the series.
Worked Example: A Finite Integral but a Divergent Sampled Series
We construct a positive continuous function whose improper integral converges even though the series of its integer samples diverges. For each \(k\ge1\), let \(w_k=1/(8k^3)\), and define a triangular spike centered at \(2k\), with height \(1/(2k)\) and support from \(2k-w_k\) to \(2k+w_k\). Set \(f(x)=1/x^2\) plus the sum of these spikes for \(x\ge1\). The supports do not overlap, and only finitely many meet any bounded interval, so \(f\) is continuous and positive.
The area of the \(k\)-th triangle is half its base times its height:
Consequently, the total integral is bounded by the integral of \(1/x^2\) plus the sum of the triangle areas:
At each even integer, the spike reaches its height, so
Thus the even-indexed terms alone have sum at least \(\frac12\sum_{k=1}^\infty 1/k\), which diverges. Therefore \(\sum_{n=1}^\infty f(n)\) diverges. The function is not nonincreasing: at every spike its value rises from the left endpoint to the center. The Integral Test cannot be used to equate the two outcomes. The direct calculation checks the discrepancy: the integral is finite, while the sampled series diverges.
A common invalid argument is, “The function is positive and continuous, and its integral converges, so the series converges.” The example shows why the omitted monotonicity condition matters: narrow spikes can contribute little area but large values at selected integers.
Direct Comparison: The Inequality Direction Is Essential
For nonnegative terms, a smaller series is controlled by a larger one only in the appropriate direction. To prove convergence by comparison, bound the terms above by those of a known convergent series. To prove divergence, bound them below by those of a known divergent series.
The reason is visible in the partial sums. If \(0\le a_n\le b_n\), then for every \(m\ge N\),
If the partial sums on the right are bounded because \(\sum b_n\) converges, the increasing partial sums on the left are bounded as well and therefore converge. Reversing the inequality does not give this bound. In the divergence case, if \(0\le b_n\le a_n\), the partial sums of \(a_n\) are at least those of \(b_n\), so they cannot remain bounded when the latter diverge.
Worked Example: A Reversed Comparison
Consider \(\sum_{n=1}^\infty 1/n\). An invalid argument might claim \(1/n\le1/n^2\), then use convergence of the \(p\)-series \(\sum 1/n^2\) to declare the harmonic series convergent. Check the proposed inequality at \(n=2\): \(1/2\le1/4\) is false. In fact, for every \(n\ge1\),
which is the reverse direction. Since \(\sum 1/n\) is the divergent harmonic series, this inequality does not prove convergence of the harmonic series. The attempted conclusion was not supported. Checking the comparison at \(n=2\) already exposes the error, and the actual inequality confirms that the convergent series is the smaller one.
For series with terms of changing sign, direct comparison should be applied to nonnegative quantities, commonly \(\lvert a_n\rvert\), when the goal is to establish absolute convergence. An inequality between signed terms alone does not provide the partial-sum bounds used above.
Limit Comparison: A Zero Ratio Does Not Mean “Same Behavior”
The standard same-behavior conclusion for Limit Comparison requires a positive finite limiting ratio. If that limit is zero or infinite, the conclusion is directional rather than symmetric. For example, if \(a_n/b_n\to0\) for positive terms, then eventually \(a_n\le b_n\). Convergence of \(\sum b_n\) then implies convergence of \(\sum a_n\); divergence of \(\sum b_n\) does not by itself settle \(\sum a_n\).
Worked Example: The Limit Is Zero, but the Series Behave Differently
Let \(a_n=1/n^2\) and \(b_n=1/n\). Their ratio is
An incorrect use of the positive-finite version of Limit Comparison would claim that the series have the same behavior. But \(\sum 1/n^2\) converges, while \(\sum 1/n\) diverges. Directly, \(0<a_n\le1/n^2\), and the \(p\)-series with \(p=2\) converges, so \(\sum a_n\) converges. The benchmark \(\sum b_n=\sum1/n\) is the harmonic series and diverges. The computed ratio and the two benchmark conclusions verify that a zero limit does not justify the same-behavior conclusion.
Alternating Series: Terms Must Decrease Eventually
The Alternating Series Test requires both that the magnitudes tend to zero and that they are eventually nonincreasing. Neither condition replaces the other. In particular, terms tending to zero and having alternating signs do not guarantee convergence when their magnitudes repeatedly increase.
Worked Example: Alternating Terms That Tend to Zero but Diverge
Define positive magnitudes by \(b_{2k-1}=1/k\) and \(b_{2k}=1/(2k)\), and consider \(\sum_{n=1}^\infty(-1)^{n-1}b_n\). Both subsequences \(1/k\) and \(1/(2k)\) tend to zero, so \(b_n\to0\). But the magnitudes are not eventually nonincreasing: for each \(k\),
because \(k+1<2k\) for \(k>1\). Pair consecutive terms. For every \(m\ge1\), their first \(2m\) partial sum is
These even partial sums grow without bound because the harmonic series diverges. The alternating series therefore diverges, despite its terms tending to zero. The calculation verifies why the missing monotonicity condition cannot be ignored.
Ratio Test: A Limit of One Is Inconclusive
When the Ratio Test gives a limit of \(1\), it does not say that a series converges, diverges, or is “on the edge” in any more specific sense. One must use another test or argument. Also, if the required ratio limit does not exist, the standard limit form of the test provides no conclusion.
Worked Example: Two Series with Ratio Limit One
For \(a_n=1/n\), the ratio is
The Ratio Test is inconclusive. The harmonic-series benchmark establishes that \(\sum 1/n\) diverges.
For comparison, let \(c_n=1/n^2\). Then
This Ratio Test is also inconclusive, but the \(p\)-series benchmark with \(p=2\) shows that \(\sum 1/n^2\) converges. The two calculations verify that the same ratio limit of \(1\) is compatible with different outcomes.
A Brief Distinction: The Root Test
The root test is a separate test, based on the limit superior of the \(n\)-th roots of the absolute values of the terms. A value below \(1\) implies absolute convergence, and a value above \(1\) implies divergence; a value of \(1\) is inconclusive. It is not the Ratio Test, and it does not repair a failed hypothesis of another test. The root test is outside the scope of the AP-focused methods considered here.
A Reliable Diagnostic Routine
Are they eventually nonnegative when the comparison or Integral Test requires positivity? For sign-changing terms, decide whether the argument concerns the original series or absolute convergence.
For the Integral Test, verify the function is positive, continuous, nonincreasing, and satisfies \(a_n=f(n)\). For the Alternating Series Test, verify the signs alternate and the magnitudes eventually decrease to zero.
In direct comparison, make sure the inequality points the way needed. In Limit Comparison, distinguish a positive finite ratio from zero or infinity. In the Ratio Test, treat a limit of one as inconclusive.
A failed hypothesis does not prove the opposite conclusion. It means only that this test has not answered the question.
On an AP exam, writing down a test’s name is not enough: the justification should make clear that its conditions hold and that its conclusion matches the result being claimed. A careful hypothesis check often catches the error before lengthy calculations begin.
Check Your Understanding
For each question, identify the hypothesis or conclusion that needs attention.
- A positive continuous function has a convergent improper integral, but it is not eventually nonincreasing. Can the Integral Test establish convergence of its sampled series?
- To prove convergence by direct comparison, should the given nonnegative terms be bounded above or below by a known convergent series?
- If \(a_n/b_n\to0\) and \(\sum b_n\) diverges, does Limit Comparison alone determine whether \(\sum a_n\) converges?
- Do alternating signs and terms tending to zero suffice for the Alternating Series Test if the magnitudes are not eventually nonincreasing?
- What does the Ratio Test conclude if the ratio limit is \(1\)?
