What Does It Mean for a Series to Converge?
An infinite series is the sum of infinitely many terms: Σₙ₌₁^∞ aₙ = a₁ + a₂ + a₃ + …. You cannot literally add infinitely many numbers, so what does the sum mean? It is defined as the limit of the partial sums: Sₙ = a₁ + a₂ + … + aₙ.
{Sₙ} approaches a finite limit S as n → ∞, the series converges to S. If Sₙ grows without bound or oscillates, the series diverges.The series Σₙ₌₀^∞ (1/2)ⁿ = 1 + 1/2 + 1/4 + 1/8 + … has partial sums Sₙ = 2 − (1/2)ⁿ, which approach 2.
The series converges to 2The series Σₙ₌₁^∞ 1/n = 1 + 1/2 + 1/3 + 1/4 + … looks like it should converge — the terms shrink to zero — but it diverges to infinity.
Why? Group the terms: 1/3 + 1/4 ≥ 2·(1/4) = 1/2, then 1/5 + 1/6 + 1/7 + 1/8 ≥ 4·(1/8) = 1/2, and so on.
You can pile up infinitely many half-units, so the sum is infinite.
This single example destroys the most common student intuition: terms going to zero is necessary for convergence, but not sufficient.
Why we need tests
For most series, you cannot find a closed form for Sₙ — and therefore cannot compute the limit directly. The convergence tests in this guide give you a way to decide whether the series converges without evaluating the sum. They tell you the verdict (converges / diverges), and sometimes a bound on the error — but most of them do not give you the value S.
That is the deal you accept when working with series: you can usually decide convergence cheaply, but evaluating the actual sum is much harder and often impossible in closed form.
The Divergence Test (nth-Term Test) — Your First Check
The cheapest test is the Divergence Test, also called the nth-Term Test.
lim_{n→∞} aₙ ≠ 0 (including the case where the limit does not exist), then Σ aₙ diverges.The contrapositive direction is important: if the terms do go to zero, this test tells you nothing. The harmonic series above proves it — 1/n → 0 yet the sum diverges. So the divergence test can only ever kill a series; it can never confirm convergence.
How to use it
Always run the divergence test first. It is a one-line check, and it eliminates the easy cases.
Does Σ n/(n+1) converge?
Compute lim_{n→∞} n/(n+1) = 1 ≠ 0.
DivergesDoes Σ cos(n) converge?
cos(n) oscillates between −1 and 1 and has no limit.
DivergesDoes Σ (−1)ⁿ converge?
Terms 1, −1, 1, −1, … have no limit.
DivergesDoes Σ 1/n² converge?
Terms 1/n² → 0. The divergence test says nothing. You must use another test (it does converge — see p-series below).
Rule of thumb. Spend five seconds checking the limit. If it is non-zero, you are done. If it is zero, move on to a real test. Never claim convergence based on terms-going-to-zero alone.
The Two Benchmark Families — Geometric Series and p-Series
Two families of series have closed-form convergence rules and serve as benchmarks for every other test.
Σₙ₌₀^∞ a·rⁿ = a + ar + ar² + ar³ + …. This series converges if and only if |r| < 1. When it converges, the sum is a / (1 − r).Σ (1/3)ⁿ converges (|r| = 1/3 < 1), sum = 1/(1 − 1/3) = 3/2.
Σ (5/4)ⁿ diverges (|r| = 5/4 > 1).
Σ (−1/2)ⁿ converges, sum = 1/(1 − (−1/2)) = 2/3.
Geometric series are the only common family where you can both decide convergence and get the sum in closed form.
Σₙ₌₁^∞ 1/nᵖ converges if and only if p > 1.Notes: p = 1 is the harmonic series (divergent). p = 2 gives Σ 1/n² = π²/6 (the Basel problem). p = 1/2 is Σ 1/√n (divergent). The cutoff is sharp: p = 1.0001 converges, p = 1 diverges.
Why p-series matter so much. They are the workhorse comparison standard. When you see a series whose terms look like 1/n^p for some power, you almost always finish with comparison or limit comparison against the matching p-series.
Memorize the rules. Geometric: converges iff |r| < 1. p-series: converges iff p > 1. Every other test reduces to comparing against one of these two.
The Integral Test — When Series Behave Like Integrals
The Integral Test connects a series to an improper integral.
f(x) be a continuous, positive, decreasing function on [N, ∞) with f(n) = aₙ. Then Σₙ₌N^∞ aₙ and ∫_N^∞ f(x) dx either both converge or both diverge.The three required hypotheses
- Continuous
So the integral is defined.
- Positive
Series with negative or oscillating terms need a different test.
- Eventually decreasing
The comparison between rectangles and the curve breaks if
fbounces around.
If any one fails, do not apply this test.
The intuition. Draw the histogram of rectangles with width 1 and heights aₙ = f(n). For a decreasing f, each rectangle sits between two pieces of the curve, so the sum and the integral differ by a bounded amount — they share the same convergence verdict.
The Integral Test gives the entire p-series rule for free.
For p ≠ 1: ∫_1^∞ x^{−p} dx = [x^{1−p}/(1−p)]_1^∞.
If p > 1, 1 − p < 0, so x^{1−p} → 0 and the integral evaluates to 1/(p−1), finite — so Σ 1/nᵖ converges.
If p < 1, x^{1−p} → ∞, the integral diverges, and so does the series.
For p = 1: ∫_1^∞ 1/x dx = lim_{b→∞} ln(b) = ∞, divergent.
Does Σₙ₌₂^∞ 1/(n·ln(n)) converge?
The function f(x) = 1/(x·ln(x)) is continuous, positive, and decreasing on [2, ∞).
Compute the integral with u = ln(x): ∫_2^∞ 1/(x·ln(x)) dx = ∫_{ln(2)}^∞ 1/u du = ∞.
Divergent — and so is the seriesWhat the integral test does not give you. It does not give the value of the sum even when both converge. The sum and the integral differ by a positive amount (which leads to the integral-test remainder estimate, useful for bounding partial-sum error).
Comparison and Limit Comparison — Beating a Series Against a Known One
Many series do not yield to a direct integral but compare favorably to a known geometric or p-series.
0 ≤ aₙ ≤ bₙ for all large n. (1) If Σ bₙ converges, then Σ aₙ converges. (2) If Σ aₙ diverges, then Σ bₙ diverges.The two valid directions. Bigger-than-divergent → divergent. Smaller-than-convergent → convergent. The other two combinations tell you nothing.
Does Σ 1/(n² + n) converge?
Compare to Σ 1/n², which converges (p = 2).
Since 1/(n² + n) ≤ 1/n² for all n ≥ 1, our series is bounded above by a convergent series.
ConvergentDoes Σ 1/(n − 1/2) converge?
For n ≥ 1, 1/(n − 1/2) ≥ 1/n.
Since Σ 1/n diverges (harmonic), our larger series also diverges.
Divergentaₙ, bₙ > 0 and lim_{n→∞} aₙ/bₙ = L with 0 < L < ∞. Then Σ aₙ and Σ bₙ share a verdict — both converge or both diverge.How to pick bₙ. Look at the dominant terms in the numerator and denominator of aₙ for large n; bₙ is the simplified leading behavior.
Does Σ (2n + 3)/(n³ − n + 5) converge?
Leading behavior is 2n/n³ = 2/n², so compare to bₙ = 1/n².
lim aₙ/bₙ = lim (2n + 3)·n²/(n³ − n + 5) = 2, finite and positive.
Σ 1/n² converges, so our series converges.
ConvergentDoes Σ 1/√(n² + 1) converge?
Compare to bₙ = 1/n.
lim aₙ/bₙ = lim n/√(n² + 1) = 1, finite and positive.
Σ 1/n diverges, so our series diverges.
DivergentWhy limit comparison wins in practice. Direct comparison requires producing a clean inequality, which gets messy with mixed terms. Limit comparison only requires identifying the leading behavior — much easier.
The Ratio Test and the Root Test
For series with factorials, exponentials, or powers of n, the ratio and root tests are the workhorses.
L = lim_{n→∞} |aₙ₊₁/aₙ|. (1) If L < 1, Σ aₙ converges absolutely. (2) If L > 1 (including L = ∞), Σ aₙ diverges. (3) If L = 1, the test is inconclusive — try something else.Why it works. If the ratio of successive terms approaches L < 1, the series is eventually bounded above by a convergent geometric series with ratio just slightly bigger than L.
When to use it. Use the ratio test whenever aₙ contains n!, rⁿ for a constant r, or nᵏ · cⁿ mixtures — anything where the ratio simplifies nicely.
Does Σ 2ⁿ/n! converge?
aₙ₊₁/aₙ = (2ⁿ⁺¹/(n+1)!) · (n!/2ⁿ) = 2/(n+1) → 0 = L < 1.
Converges absolutelyDoes Σ n²/2ⁿ converge?
aₙ₊₁/aₙ = ((n+1)²/2ⁿ⁺¹) · (2ⁿ/n²) = (1/2)·(1 + 1/n)² → 1/2 < 1.
ConvergesDoes Σ n!·2ⁿ/n^n converge?
aₙ₊₁/aₙ = ((n+1)!·2ⁿ⁺¹/(n+1)^{n+1}) · (n^n/(n!·2ⁿ)) = 2·(n/(n+1))^n → 2/e ≈ 0.736 < 1.
ConvergesFor Σ 1/n²: aₙ₊₁/aₙ = n²/(n+1)² → 1.
Inconclusive. Use the p-series rule instead.
L = lim_{n→∞} |aₙ|^{1/n}. Same trichotomy: L < 1 converges absolutely, L > 1 diverges, L = 1 inconclusive.When to use it. Use the root test when aₙ contains an expression raised to the nth power — e.g. (stuff)ⁿ. The nth root cleanly cancels the exponent.
Does Σ (n/(2n+1))ⁿ converge?
|aₙ|^{1/n} = n/(2n+1) → 1/2 < 1.
ConvergesDoes Σ ((1 + 1/n)^{n²})/eⁿ converge?
|aₙ|^{1/n} = (1 + 1/n)^n/e → e/e = 1.
Inconclusive — try another approach (here, careful asymptotic analysis shows divergence, but the root test itself is silent).
Important fact. If both tests apply, they always give the same value of L — so they never disagree. The root test is strictly more powerful in theory but the ratio test is usually easier to compute.
Alternating Series, Absolute vs Conditional Convergence, and a Decision Tree
Σ (−1)ⁿ bₙ or Σ (−1)ⁿ⁺¹ bₙ with bₙ > 0. If (1) bₙ is eventually decreasing and (2) lim bₙ = 0, then the series converges.Σ (−1)ⁿ⁺¹/n = 1 − 1/2 + 1/3 − 1/4 + … is the alternating harmonic series.
1/n decreases to 0, so the series converges (its sum is ln 2).
Useful error bound. For an alternating series satisfying the Leibniz hypotheses, the error after n terms is at most the first omitted term: |S − Sₙ| ≤ bₙ₊₁. This is the cleanest error estimate in all of series theory.
Absolute vs Conditional Convergence
Σ aₙ converges absolutely if Σ |aₙ| converges. It converges conditionally if Σ aₙ converges but Σ |aₙ| diverges.Why the distinction matters. Absolutely convergent series behave like finite sums: you can rearrange the terms freely and get the same value. Conditionally convergent series can be rearranged to sum to any real number you want (Riemann rearrangement theorem).
Σ (−1)ⁿ⁺¹/n converges to ln 2, but Σ |(−1)ⁿ⁺¹/n| = Σ 1/n diverges.
So the sum depends on the order.
Σ (−1)ⁿ/n²: the absolute series Σ 1/n² converges (p = 2), so the original converges absolutely.
Quick decision tree
- Compute lim aₙ
If non-zero, divergent — done.
- Recognize the form
Geometric (
rⁿ) or p-series (1/nᵖ) — apply the rule directly. - Alternating with positive part bₙ
Try Leibniz.
- Factorials, exponentials, or (stuff)ⁿ
Try the ratio or root test.
- Looks like a known p-series or geometric series with extra clutter
Try direct or limit comparison.
- Continuous, positive, decreasing
Try the integral test.
Common mistakes
Concluding convergence from aₙ → 0 alone.
Applying the integral test to a series with negative terms.
Forgetting to check that bₙ is decreasing in the alternating series test (some textbook problems exploit this).
Using direct comparison in the wrong direction — bigger-than-convergent and smaller-than-divergent give you nothing.
Confusing the ratio test's L = 1 with convergence — it means the test gave up, not that the series converges.
Claiming a conditionally convergent series can be rearranged — yes, but you change the sum if you do.
Does Σ (−1)ⁿ · n/(n² + 1) converge, and if so absolutely or conditionally?
Step 1: divergence test. aₙ → 0, inconclusive.
Step 2: absolute series Σ n/(n² + 1). Limit-compare to 1/n: lim (n·n)/(n² + 1) = 1, finite and positive. Σ 1/n diverges, so absolute series diverges. Not absolutely convergent.
Step 3: alternating series test. bₙ = n/(n² + 1); check bₙ → 0: n/(n² + 1) → 0 ✓. Check decreasing: derivative of x/(x² + 1) is (1 − x²)/(x² + 1)², negative for x > 1 ✓. So Leibniz applies — the original converges.
Conditionally convergentWhen working through convergence problems for AP Calc BC, Calc 2 finals, or a real analysis course, scan them with Solver AI for the full setup — which test to choose, every step of the limit, and the absolute-vs-conditional verdict.