Every nightclub worth its name has a bouncer at the door. The bouncer does not decide who has a good night — that happens inside. The bouncer's only job is to turn away, in about two seconds, the people who obviously have no business getting in. The divergence test (also called the nth-term test) is the bouncer for infinite series: it is the cheapest check in the whole book, and it can only ever reject — it never admits anyone.

The check is this: look at the terms a_n of the series \sum_{n=1}^{\infty} a_n and ask whether they even bother to shrink to zero. You are adding up infinitely many numbers. If, far down the line, each new number you throw on the pile still has some fixed size — each one adds at least (say) 0.4, forever — the running total cannot possibly settle down. It gets shoved by a definite amount at every single step, infinitely often. No finite total survives that.

So before you reach for any of the serious machinery — comparison, integral, ratio — you take the limit of the terms. Costs one line. If that limit is anything other than 0, or fails to exist at all, you are done: the series diverges, case closed. If the limit is zero… the bouncer shrugs and waves you through to the real interview. That second half of the story — what the test cannot tell you — is where this page will spend a lot of its energy, because it is the single most-failed exam point in the entire chapter.

The test, stated properly

The honest mathematical fact underneath the test is a statement about convergent series: convergence forces the terms to die out.

Notice this is phrased the "wrong way round" for practical use — it starts from convergence, which is exactly the thing you don't yet know. The usable tool is its contrapositive. Any implication "if P then Q" is logically identical to "if not-Q then not-P": if convergence guarantees vanishing terms, then non-vanishing terms rule out convergence.

Note the parenthesis carefully: "the limit fails to exist" counts as failing the test just as hard as a non-zero limit does. Terms that oscillate forever, or blow up to infinity, never had a limit of zero — so the series they belong to diverges.

One-directional, always. Terms going to zero is necessary for convergence, but nowhere near sufficient. The arrow only runs one way:

Bouncer logic, exactly: being sober is necessary to get into the club, but showing up sober doesn't mean you're on the guest list.

Why convergence forces the terms to zero

The whole proof is a single clean cancellation — one of the shortest genuinely useful proofs in analysis. Write s_n = a_1 + a_2 + \dots + a_n for the partial sum, and suppose the series converges to a finite limit L, meaning s_n \to L. The intuition first: if the running totals are settling onto L, then consecutive totals s_{n-1} and s_n are both squeezed up against the same number — so the gap between them is being crushed to nothing. But that gap is the term a_n. Now the same thing carefully:

Step 1 — recover a single term from neighbouring partial sums. Adding one more term takes s_{n-1} to s_n, so the difference is exactly the term itself:

Step 2 — take the limit of both sides. Let n \to \infty. The left side becomes \lim_{n\to\infty} a_n; the right side is a limit of a difference:

Step 3 — split the limit of the difference. Both pieces converge, so the limit distributes across the subtraction:

Step 4 — both partial-sum limits are the same number. As n \to \infty, the index n-1 also runs off to infinity, so s_{n-1} chases the very same limit L:

Step 5 — collect. The two copies of L annihilate:

So convergence forces the terms to zero. Contrapose it and you have the test: terms that do not vanish guarantee a divergent series. Notice what the proof did not say: it never claimed that small terms make the sum settle. It only squeezed the terms using a settlement we assumed. That asymmetry is the entire personality of this test.

Three series walk up to the door

Here is the test doing its actual day job. Three candidates, three different outcomes — and only two of them get a verdict.

Example 1 — rejected on sight. Consider

Take the limit of the terms: divide top and bottom by n,

Done. Far down the series, every new term adds roughly \tfrac12, so the partial sums climb like \tfrac{n}{2} — off to infinity. The series diverges by the divergence test, and it took one line. This is the test's home turf: any series whose terms are a ratio of same-degree polynomials fails it, because the term limit is the (non-zero) ratio of leading coefficients.

Example 2 — the test says nothing; hand it off. Now try the famous harmonic series,

The terms do go to zero, so the divergence test is stone silent. Silent — not approving. You may not write "terms \to 0, therefore converges"; that sentence has ended more exam scripts than any other in this chapter. The correct move is a hand-off to a sharper tool: the integral test or a comparison shows that the harmonic series in fact diverges — the terms shrink, just not fast enough. The vignette below has the beautiful 650-year-old proof.

Example 3 — rejected for refusing to make up its mind. Finally,

The even-indexed terms creep up towards +1 and the odd-indexed terms creep down towards -1, so \lim_{n\to\infty} a_n does not exist. That is a failure of the test's condition just as surely as a non-zero limit: the series diverges. Its partial sums swing back and forth forever between two separated clusters and never settle. Don't be fooled by the alternating signs into reaching for the alternating series test — that test needs terms shrinking to zero too, and these never do.

Watch a series fail the test

The two charts below are the divergence test made visible. Take terms a_n = L + \tfrac{1}{n}, which settle towards a chosen limit L — the dashed line in the first chart. As long as L \ne 0 the terms hover above that non-zero floor and never die out, and the second chart shows the consequence: the partial sums s_n = \sum_{k=1}^{n} a_k climb away roughly like the straight line L\,n, gaining about L with every step. A steeper floor means a faster escape — that is the divergence test's whole argument, drawn.

Now slide L down to 0 and watch the moral of this page play out. The terms now vanish, the runaway line loses its slope — and yet the partial sums still creep upward without bound, because what is left is exactly the harmonic series. The first chart passing the terms-go-to-zero check tells you nothing about whether the second chart levels off. Only the partial sums know.

See it explained