Infinite Series — Convergence Tests — Mathematics

⚡ Key Takeaways

Convergence tests are systematic criteria for deciding whether an infinite series $\sum a_n$ converges (sums to a finite value) or diverges. Key result in words: a series converges if and only if its sequence of partial sums has a finite limit, and a battery of comparison-, ratio-, root-, and integral-based tests provides practical ways to detect this without computing the limit directly. They apply throughout analysis, physics, numerical computation, and the theory of functions. Developed by Cauchy, d'Alembert, Abel, Dirichlet, Raabe, and others, c. 1700–1900.

💡 The Core Insight

A series is nothing but a sequence in disguise — the sequence of its partial sums. All convergence tests are devices for deciding whether $(S_N)$ is Cauchy without evaluating $\lim S_N$ . They do this by comparing the unknown series to a known benchmark (geometric series, $p$ -series, an integral) either directly, or asymptotically through ratios and roots. The deepest dividing line is between absolute convergence (robust, behaves like finite arithmetic) and conditional convergence (fragile, order-dependent).

DefinitionSeries, Partial Sums, Convergence

Given a sequence $(a_n)_{n\ge 1}$ of real (or complex) numbers, the $N$ -th partial sum is

1

\quad S_N = \sum_{n=1}^{N} a_n = a_1 + a_2 + \cdots + a_N.

Derivation hint: a finite sum, always well-defined; the series is the sequence $(S_N)$ . The infinite series $\sum_{n=1}^\infty a_n$ converges to $S$ if

2

\quad S = \lim_{N\to\infty} S_N, \qquad \text{i.e. } \forall \varepsilon>0\ \exists M:\ N\ge M \Rightarrow |S_N - S| < \varepsilon.

Derivation hint: ordinary sequence-limit definition applied to the partial sums. If no finite limit exists, the series diverges. The series converges absolutely if $\sum |a_n|$ converges, and converges conditionally if $\sum a_n$ converges but $\sum |a_n|$ diverges.

TheoremCauchy Criterion for Series

$\sum a_n$ converges if and only if for every $\varepsilon > 0$ there is $M$ such that for all $n > m \ge M$ ,

3

\quad \left| \sum_{k=m+1}^{n} a_k \right| < \varepsilon.

Derivation hint: $(S_N)$ converges iff it is Cauchy, by completeness of $\mathbb{R}$ ; the tail sum is $S_n - S_m$ .

TheoremThe Standard Benchmarks

Two reference series anchor all comparison-based tests.

Geometric series: for $r \in \mathbb{R}$ ,

4

\quad \sum_{n=0}^{\infty} r^n = \frac{1}{1-r} \ \text{ if } |r|<1, \quad \text{diverges if } |r|\ge 1.

Derivation hint: $S_N(1-r) = 1 - r^{N+1}$ ; let $N\to\infty$ with $|r|<1$ .

$p$ -series: for $p \in \mathbb{R}$ ,

5

\quad \sum_{n=1}^{\infty} \frac{1}{n^p} \ \text{ converges } \iff p > 1.

Derivation hint: compare with $\int_1^\infty x^{-p}\,dx$ , which is finite iff $p>1$ (see Integral Test).

TheoremCatalogue of Convergence Tests

Let $a_n$ denote the terms; assume series start at a convenient index. 1. $n$ -th Term (Divergence) Test. If $a_n \not\to 0$ , then $\sum a_n$ diverges. (Necessary, not sufficient.) 2. Comparison Test. If $0 \le a_n \le b_n$ and $\sum b_n$ converges, then $\sum a_n$ converges. If $0 \le b_n \le a_n$ and $\sum b_n$ diverges, so does $\sum a_n$ . 3. Limit Comparison Test. If $a_n, b_n > 0$ and $\lim a_n/b_n = L \in (0,\infty)$ , the two series share their fate. 4. Integral Test. If $f$ is positive, continuous, decreasing with $f(n) = a_n$ , then $\sum a_n$ and $\int_1^\infty f$ converge or diverge together. 5. Ratio Test (d'Alembert). Let $L = \lim |a_{n+1}/a_n|$ . Then $L<1\Rightarrow$ absolute convergence, $L>1\Rightarrow$ divergence, $L=1\Rightarrow$ inconclusive. 6. Root Test (Cauchy). Let $L = \limsup |a_n|^{1/n}$ . Same trichotomy as the ratio test. 7. Alternating Series Test (Leibniz). If $b_n \downarrow 0$ , then $\sum (-1)^n b_n$ converges, with error bounded by the first omitted term. 8. Dirichlet's Test. If partial sums of $\sum a_n$ are bounded and $b_n \downarrow 0$ , then $\sum a_n b_n$ converges. 9. Abel's Test. If $\sum a_n$ converges and $(b_n)$ is monotone and bounded, then $\sum a_n b_n$ converges. 10. Raabe's Test. Let $R = \lim n\!\left(\frac{a_n}{a_{n+1}} - 1\right)$ . Then $R>1\Rightarrow$ convergence, $R<1\Rightarrow$ divergence (refines the ratio test at $L=1$ ).

ProofIntegral Test — the Master Comparison

We prove the Integral Test, from which the $p$ -series criterion follows.

Step 1. Let $f:[1,\infty)\to[0,\infty)$ be continuous, positive, decreasing, with $f(n)=a_n$ . Because $f$ decreases, on each interval $[k, k+1]$ ,

6

\quad f(k+1) \le \int_k^{k+1} f(x)\,dx \le f(k).

Derivation hint: a decreasing function is bounded below by its right endpoint value and above by its left endpoint value on the interval.

Step 2. Sum [6] from $k=1$ to $N-1$ :

7

\quad \sum_{k=2}^{N} a_k \le \int_1^{N} f(x)\,dx \le \sum_{k=1}^{N-1} a_k.

Derivation hint: telescoping the integral $\int_1^N = \sum_k \int_k^{k+1}$ and re-indexing the term sums.

Step 3. If $\int_1^\infty f < \infty$ , the right inequality bounds the partial sums of nonnegative terms; a bounded increasing sequence converges, so $\sum a_n$ converges. If $\int_1^\infty f = \infty$ , the left inequality forces $S_N \to \infty$ . $\blacksquare$

Corollary ( $p$ -series). With $f(x)=x^{-p}$ , $\int_1^\infty x^{-p}\,dx = \frac{1}{p-1}$ if $p>1$ and diverges if $p\le 1$ , giving [5].

∎

ProofRatio Test from the Geometric Series

Step 1. Suppose $L = \lim |a_{n+1}/a_n| < 1$ . Pick $r$ with $L < r < 1$ . There is $M$ with

8

\quad \left|\frac{a_{n+1}}{a_n}\right| < r \quad \text{for all } n \ge M.

Derivation hint: definition of limit with $\varepsilon = r - L > 0$ .

Step 2. Iterating [8], $|a_{M+k}| < r^k |a_M|$ , so

9

\quad \sum_{n=M}^{\infty} |a_n| \le |a_M| \sum_{k=0}^{\infty} r^k = \frac{|a_M|}{1-r} < \infty.

Derivation hint: geometric domination, then sum the geometric series [4]. By comparison with this convergent geometric tail, $\sum|a_n|$ converges, so $\sum a_n$ converges absolutely. If $L>1$ , eventually $|a_{n+1}|>|a_n|$ , so $a_n\not\to 0$ and the series diverges by the $n$ -th term test. $\blacksquare$

∎

ProofAlternating Series Test

Let $b_n \downarrow 0$ and $S_N = \sum_{n=1}^N (-1)^{n-1} b_n$ .

Step 1. The even partial sums increase and the odd ones decrease:

10

\quad S_{2k} = S_{2k-2} + (b_{2k-1}-b_{2k}) \ge S_{2k-2}, \qquad S_{2k+1} = S_{2k-1} - (b_{2k}-b_{2k+1}) \le S_{2k-1}.

Derivation hint: group consecutive terms in pairs; each pair has the sign forced by monotonicity $b_n \ge b_{n+1}$ .

Step 2. Since $S_{2k} \le S_{2k+1}$ and the even sequence increases while bounded above (by any odd sum), both subsequences converge; and $S_{2k+1}-S_{2k}=b_{2k+1}\to 0$ , so they share a common limit $S$ . Hence $S_N \to S$ , and $|S - S_N| \le b_{N+1}$ . $\blacksquare$

∎

mathematician · Augustin-Louis Cauchy | 1789–1857

Cauchy gave the first rigorous definition of convergence via partial sums and the Cauchy criterion in his 1821 Cours d'analyse, ending a century in which series were manipulated formally and divergent sums (like $1-1+1-\cdots$ ) were assigned values by analogy. He proved the root test and the condensation test that bears his name, and insisted that a series be shown to converge before its sum is used — a methodological revolution. His near-contemporary Jean le Rond d'Alembert (1717–1783) had earlier introduced the ratio test in the context of binomial and hypergeometric series.

🧠 Intuition

Picture each term $a_n$ as the length of a step you take. Convergence asks whether the total walking distance is finite. For the series to even have a chance, the steps must shrink to zero ( $n$ -th term test) — but shrinking is not enough: the harmonic steps $1, \tfrac12, \tfrac13, \dots$ shrink yet you still walk infinitely far, because they shrink too slowly. The boundary is sharp at $p=1$ : at $1/n^p$ , powers above $1$ shrink fast enough, powers at or below $1$ do not. The ratio and root tests ask a different question: how fast do the terms shrink, relative to a geometric series? If each step is reliably less than $r<1$ times the previous one, your path is dominated by a geometric series of finite length. When the shrinkage is exactly geometric in the limit ( $L=1$ ), the tests go blind — that knife-edge contains the entire family $1/n^p$ , where the ratio tends to $1$ regardless of $p$ .

Alternating series convergence is a cancellation phenomenon: each step partly undoes the last, so even slowly-shrinking steps (like $1/n$ ) net out to a finite displacement. This is why $\sum (-1)^{n-1}/n = \ln 2$ converges while $\sum 1/n$ diverges — the same magnitudes, but with signs that cancel.

Figure

The convergence landscape of

\sum 1/n^p

: the partial-sum staircase rises without bound for

p\le 1

and levels off for

p>1

.

```svg

```

ExampleRatio Test on a Factorial Series

Given: Determine convergence of $\displaystyle \sum_{n=1}^{\infty} \frac{3^n}{n!}$ .

Step 1. Form the ratio of consecutive terms:

\frac{a_{n+1}}{a_n} = \frac{3^{n+1}/(n+1)!}{3^n/n!} = \frac{3^{n+1}}{3^n}\cdot\frac{n!}{(n+1)!} = \frac{3}{n+1}.

Step 2. Take the limit:

L = \lim_{n\to\infty} \frac{3}{n+1} = 0.

Step 3. Since $L = 0 < 1$ , the Ratio Test gives absolute convergence.

Bonus: the sum is in fact $\sum_{n=0}^\infty 3^n/n! = e^3$ , so $\sum_{n=1}^\infty = e^3 - 1 \approx 19.086$ (dimensionless).

\boxed{\text{Converges absolutely; } L = 0 < 1,\quad S = e^3 - 1 \approx 19.086.}

ExampleIntegral Test and a

p

-Series Near the Boundary

Given: Does $\displaystyle \sum_{n=2}^{\infty} \frac{1}{n (\ln n)^p}$ converge? (Take general $p$ , then $p = 1$ and $p = 2$ .)

Step 1. Let $f(x) = \frac{1}{x(\ln x)^p}$ — positive, continuous, decreasing for $x\ge 2$ . Substitute $u = \ln x$ , $du = dx/x$ :

\int_2^\infty \frac{dx}{x(\ln x)^p} = \int_{\ln 2}^{\infty} \frac{du}{u^p}.

Step 2. This is a $u$ -integral of power $-p$ , finite iff $p > 1$ . Hence the series converges iff $p>1$ .

Step 3 (specific values).

$p = 1$ : integral $= \int_{\ln 2}^\infty du/u = \infty$ → diverges (a famous borderline failure).
$p = 2$ : integral $= \left[-1/u\right]_{\ln 2}^\infty = 1/\ln 2 \approx 1.443$ → converges.

\boxed{\text{Converges} \iff p > 1;\quad p=1 \text{ diverges},\ p=2 \text{ converges } (\text{integral} \approx 1.443).}

⚠️ Common mistake

"If $a_n \to 0$ , the series converges." False. Vanishing terms are necessary but not sufficient. The harmonic series $\sum 1/n$ has $a_n\to 0$ yet diverges. The $n$ -th term test can only prove divergence (when $a_n \not\to 0$ ); it can never prove convergence.

"The Ratio Test settled it; it's never inconclusive." When $L = 1$ the Ratio Test says nothing — and this case is huge: every $p$ -series has $\lim a_{n+1}/a_n = 1$ . Switch to the Integral, Raabe's, or Comparison tests there.

"I can rearrange the terms freely." Only for absolutely convergent series. A conditionally convergent series can be rearranged to sum to any chosen real number, or to $\pm\infty$ (Riemann rearrangement theorem).

"The Root and Ratio tests always agree." They have the same threshold, but the Root Test is strictly stronger: whenever the ratio limit exists the root limit equals it, but the root test (using $\limsup$ ) succeeds on some series where the ratio limit fails to exist.

CounterexampleVanishing Terms, Divergent Series

The harmonic series $\sum_{n=1}^\infty 1/n$ has terms $\to 0$ but diverges. Cauchy's condensation / grouping shows why:

\underbrace{1}_{} + \underbrace{\tfrac12}_{\ge 1/2} + \underbrace{\tfrac13+\tfrac14}_{\ge 1/2} + \underbrace{\tfrac15+\cdots+\tfrac18}_{\ge 1/2} + \cdots

Each bracketed block sums to at least $1/2$ , so the partial sums exceed $1 + \tfrac{k}{2}\to\infty$ . The terms shrink, but never fast enough.

CounterexampleRatio Test Blindness

For both $\sum 1/n$ (diverges) and $\sum 1/n^2$ (converges) the ratio limit is identical:

\lim_{n\to\infty}\frac{1/(n+1)^p}{1/n^p} = \lim_{n\to\infty}\left(\frac{n}{n+1}\right)^p = 1 \quad \text{for every } p.

The Ratio Test cannot distinguish them — proof that $L=1$ is genuinely undecided and requires a finer instrument.

connection

Convergence tests are the gateway to several deeper structures. The radius of convergence of a Power Series is computed directly by the Root Test via the Cauchy–Hadamard formula $1/R = \limsup |c_n|^{1/n}$ . The distinction between absolute and conditional convergence governs when one may interchange limits, integrate, or differentiate term-by-term — the heart of Uniform Convergence and the Weierstrass $M$ -test. The Integral Test forges the precise link with Improper Integrals, so that knowing one settles the other. Beyond convergence lies the theory of Summation Methods (Cesàro, Abel, Borel), which assigns meaningful values to certain divergent series. Finally, Taylor Series and Fourier Series are the principal sources of series whose convergence one must test in practice.

🔬 Where this appears

⏳ History

- c. 1350 — Oresme. Nicole Oresme proves the harmonic series diverges by the grouping argument, centuries ahead of its time.

- 1668 — Mercator/Gregory.

\ln(1+x)

series; early systematic use of infinite series for functions.

- 1689 — Jakob Bernoulli. Tractatus de seriebus infinitis; rediscovers harmonic divergence, studies

\sum 1/n^2

(the "Basel problem").

- 1734–35 — Euler. Solves the Basel problem:

\sum 1/n^2 = \pi^2/6

; brilliant but formal manipulation of divergent and convergent series alike.

- c. 1750 — d'Alembert. Introduces the ratio test for binomial/hypergeometric series.

- 1821 — Cauchy. Cours d'analyse sets convergence on rigorous foundations: partial-sum definition, Cauchy criterion, root test, condensation test.

- 1826 — Abel. Proves Abel's theorem and the summation-by-parts identity; clarifies behavior at the boundary of the disk of convergence.

- 1827 — Raabe. Publishes Raabe's test for the

L=1

borderline of the ratio test.

- 1837 — Dirichlet. Dirichlet's test and rearrangement insights; foundational work linking series and Fourier analysis.

- 1854 — Riemann. Proves the rearrangement theorem: a conditionally convergent series can be reordered to any sum.

open-problem

The classical tests are not exhaustive: between the Ratio Test ( $L=1$ ) and Raabe's Test ( $R=1$ ) lies an infinite hierarchy of ever-finer criteria (the logarithmic / Kummer / Bertrand–Gauss tests), and Kummer's theorem shows no single test can decide every case — for any test there is a series it cannot resolve. A practical frontier is divergent-series summation: assigning consistent values to series like the QFT perturbation expansions, where convergence fails but the series is "asymptotic" to a true answer; whether such expansions are Borel-summable and resurgence-theoretic methods recover the exact result are active research questions. In number theory, the convergence boundary $\Re(s)=1$ of $\sum n^{-s}$ analytically continues to a function whose nontrivial zeros are the subject of the Riemann Hypothesis — an open problem at the very edge of "convergence."

Infinite Series — Convergence Tests

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also