Power Series and Convergence

⚡ Key Takeaways

A power series is an infinite series of the form $\sum_{n=0}^\infty a_n (x-c)^n$ . Each such series converges on an interval centered at $c$ — possibly a single point, possibly the whole line — governed by a single number called the radius of convergence $R$ . Within that interval the series defines an infinitely differentiable function that can be integrated and differentiated term by term. Power series are the backbone of analytic function theory, Taylor expansions, and the rigorous construction of $e^x$ , $\sin x$ , $\cos x$ , and the solutions of countless differential equations.

💡 The Core Insight

The convergence of $\sum a_n(x-c)^n$ at one point forces its behavior everywhere closer to the center. If the series converges at a point $x_0$ , then the terms $a_n(x_0-c)^n$ are bounded, and for any $x$ with $|x-c|<|x_0-c|$ the tail is dominated by a geometric series with ratio less than 1. Thus a single convergence point propagates convergence inward — and this geometric domination is the engine behind every major theorem on power series: the existence of $R$ , absolute convergence, uniform convergence on compact subdisks, and term-by-term calculus.

DefinitionPower Series

A power series centered at $c$ (or about $c$ ) is a series

1

\quad \sum_{n=0}^{\infty} a_n (x-c)^n = a_0 + a_1(x-c) + a_2(x-c)^2 + \cdots

Derivation hint: this is a definition; the $n=0$ term is read as $a_0$ even when $x=c$ , taking $0^0=1$ . Here:

$x$ is the variable (real or complex),
$c$ is the fixed center,
$a_n$ are the coefficients (constants), and
the series is a function of $x$ wherever it converges.

The domain of convergence is the set $\{x : \text{the series converges}\}$ . Setting $y = x-c$ reduces any power series to one centered at $0$ , so we often study $\sum a_n x^n$ without loss of generality.

DefinitionRadius of Convergence

The radius of convergence $R$ of $\sum a_n(x-c)^n$ is

2

\quad R = \sup\{\, r \ge 0 : \textstyle\sum |a_n| r^n \text{ converges} \,\} \in [0,\infty].

Derivation hint: defined as a supremum of radii of absolute convergence; equals $0$ if the set is $\{0\}$ and $\infty$ if unbounded. The interval of convergence is the set of all $x$ for which the series converges, which is one of $\{c\}$ , $(c-R,c+R)$ , $[c-R,c+R)$ , $(c-R,c+R]$ , or $[c-R,c+R]$ when $0<R<\infty$ , and all of $\mathbb{R}$ when $R=\infty$ .

TheoremCauchy–Hadamard Formula

The radius of convergence is given explicitly by

3

\quad \frac{1}{R} = \limsup_{n\to\infty} |a_n|^{1/n},

Derivation hint: apply the root test to $\sum a_n(x-c)^n$ ; the series converges absolutely when $\limsup |a_n(x-c)^n|^{1/n}<1$ . with the conventions $1/0 = \infty$ and $1/\infty = 0$ . When the limit exists, the ratio test gives the often-easier formula

4

\quad R = \lim_{n\to\infty} \left| \frac{a_n}{a_{n+1}} \right|,

Derivation hint: the ratio test on $\sum a_n(x-c)^n$ requires $\lim |a_{n+1}(x-c)^{n+1}/(a_n(x-c)^n)| = |x-c|/R < 1$ . provided this limit exists (it may fail to even when the Cauchy–Hadamard limsup does not).

TheoremConvergence Behavior — Abel's Lemma

If $\sum a_n(x-c)^n$ converges at $x=x_0 \ne c$ , then it converges absolutely for every $x$ with $|x-c| < |x_0-c|$ . Consequently the set of convergence is an interval symmetric about $c$ , and $R$ in $[2]$ is well-defined.

ProofAbel's Lemma and the Existence of R

Step 1 (Bounded terms). Suppose $\sum a_n(x_0-c)^n$ converges. The general term of a convergent series tends to $0$ , so $a_n(x_0-c)^n \to 0$ . A convergent-to-zero sequence is bounded: there is $M>0$ with

5

\quad |a_n (x_0-c)^n| \le M \quad \text{for all } n.

Derivation hint: every convergent sequence is bounded; here the limit is $0$ .

Step 2 (Geometric domination). Let $x$ satisfy $|x-c| < |x_0-c|$ and set the ratio

6

\quad t = \frac{|x-c|}{|x_0-c|} \in [0,1).

Derivation hint: $|x-c|<|x_0-c|$ makes the quotient strictly less than $1$ . Then

7

\quad |a_n(x-c)^n| = |a_n(x_0-c)^n| \cdot t^n \le M\, t^n.

Derivation hint: factor $(x-c)^n = (x_0-c)^n \cdot t^n$ (up to sign/phase) and apply bound $[5]$ .

Step 3 (Comparison). Since $0\le t<1$ , the geometric series $\sum_{n=0}^\infty M t^n = M/(1-t)$ converges. By the comparison test, $\sum |a_n(x-c)^n|$ converges, so the original series converges absolutely.

Step 4 (Existence of $R$ ). Let $S=\{r\ge 0 : \sum|a_n|r^n \text{ converges}\}$ and $R=\sup S$ . Steps 1–3 show that if $r\in S$ then $[0,r)\subseteq S$ , so $S$ is an interval starting at $0$ . For $|x-c|<R$ pick $r$ with $|x-c|<r<R$ , $r\in S$ ; convergence follows. For $|x-c|>R$ the terms cannot tend to $0$ (else by Step 1–3 a larger radius would converge), so the series diverges. $\blacksquare$

∎

ProofCauchy–Hadamard via the Root Test

Step 1. The root test examines $L(x)=\limsup_n |a_n(x-c)^n|^{1/n} = |x-c|\cdot \limsup_n |a_n|^{1/n}$ .

Step 2. Write $\alpha = \limsup_n |a_n|^{1/n}$ , so $L(x) = \alpha|x-c|$ . The root test gives absolute convergence when $L(x)<1$ , i.e. $|x-c| < 1/\alpha$ , and divergence when $L(x)>1$ , i.e. $|x-c|>1/\alpha$ .

Step 3. Comparing with Definition $[2]$ , the threshold is exactly $R = 1/\alpha$ , proving $[3]$ . The boundary case $L(x)=1$ is inconclusive — this is precisely why endpoints require separate analysis. $\blacksquare$

∎

mathematician · Niels Henrik Abel | 1802–1829

The Norwegian mathematician Abel, who died of tuberculosis at 26, gave power series their first rigorous treatment. In his 1826 paper on the binomial series he proved what is now Abel's Lemma (the inward-propagation of convergence) and the celebrated Abel's Theorem on continuity up to the boundary: if $\sum a_n$ converges, then $\sum a_n x^n \to \sum a_n$ as $x\to 1^-$ . He sharply criticized the casual eighteenth-century manipulation of divergent series — "divergent series are the invention of the devil" — and insisted that a series equation means nothing until its convergence is established. His work, alongside Cauchy's, founded the modern theory of convergence.

🧠 Intuition

Picture the power series as a tug-of-war between two forces: the coefficients $a_n$ trying to make terms large, and the powers $(x-c)^n$ trying to shrink them when $|x-c|$ is small. The radius of convergence is the exact distance from the center at which these forces balance. Inside $R$ , the powers win — terms decay geometrically and the sum is finite. Outside $R$ , the coefficients win — terms blow up and the series diverges. The quantity $|a_n|^{1/n}$ measures the "average growth rate per term" of the coefficients. If coefficients grow like $\rho^n$ , then $|a_n|^{1/n}\to\rho$ , and we need $|x-c|<1/\rho$ to overpower them. Coefficients that grow factorially ( $a_n = n!$ ) defeat any nonzero $x$ , giving $R=0$ ; coefficients that decay factorially ( $a_n=1/n!$ ) are conquered by any $x$ , giving $R=\infty$ — which is exactly why $e^x = \sum x^n/n!$ converges everywhere. The endpoints $x=c\pm R$ are the knife's edge where neither force wins outright, so anything can happen there and must be checked by hand.

Figure

Convergence structure of a power series: absolute convergence inside the radius, divergence outside, and undetermined endpoints

ExampleGeometric Series and Its Radius

Given: $\displaystyle\sum_{n=0}^\infty x^n = 1 + x + x^2 + \cdots$ , so $a_n = 1$ , center $c=0$ .

Step 1 — Radius via ratio test [4]:

8

\quad R = \lim_{n\to\infty}\left|\frac{a_n}{a_{n+1}}\right| = \lim_{n\to\infty}\frac{1}{1} = 1.

Derivation hint: all coefficients equal $1$ , so the ratio is identically $1$ .

Step 2 — Sum inside the interval: the partial sum $S_N = \frac{1-x^{N+1}}{1-x}$ , and for $|x|<1$ we have $x^{N+1}\to 0$ , so

9

\quad \sum_{n=0}^\infty x^n = \frac{1}{1-x}, \qquad |x|<1.

Derivation hint: telescoping/geometric partial-sum identity $(1-x)S_N = 1-x^{N+1}$ , then take $N\to\infty$ .

Step 3 — Endpoints: at $x=1$ , terms are $1\not\to 0$ → diverges. At $x=-1$ , terms are $(-1)^n\not\to 0$ → diverges.

\boxed{R = 1,\quad \text{interval of convergence } (-1,1),\quad \text{sum } = \tfrac{1}{1-x}}

ExampleA Series Convergent at One Endpoint Only

Given: $\displaystyle\sum_{n=1}^\infty \frac{x^n}{n}$ , so $a_n = 1/n$ , center $c=0$ .

Step 1 — Radius:

10

\quad R = \lim_{n\to\infty}\left|\frac{a_n}{a_{n+1}}\right| = \lim_{n\to\infty}\frac{1/n}{1/(n+1)} = \lim_{n\to\infty}\frac{n+1}{n} = 1.

Derivation hint: ratio of $1/n$ to $1/(n+1)$ simplifies to $(n+1)/n \to 1$ .

Step 2 — Endpoint $x=1$ : $\sum 1/n$ is the harmonic series, which diverges.

Step 3 — Endpoint $x=-1$ : $\sum (-1)^n/n$ is alternating with terms decreasing to $0$ ; by the alternating series test it converges (conditionally) to $-\ln 2$ .

\boxed{R=1,\quad \text{interval of convergence } [-1,1),\quad \text{sum }=-\ln(1-x)}

This shows the interval can be half-open: convergence at one endpoint, divergence at the other — settled only by separate tests.

⚠️ Common mistake

The ratio/root test is inconclusive at the endpoints. Both tests deliver $|x-c|/R$ as the limiting ratio, which equals exactly $1$ at $x=c\pm R$ — the inconclusive case. You must always substitute the endpoints back into the original series and apply a different test (alternating series, $p$ -series, comparison, divergence test). Never assume the endpoints are included or excluded by symmetry; Example 2 shows they can behave differently.

Convergence ≠ absolute convergence at endpoints. Inside the open interval convergence is automatically absolute (Abel's Lemma). At endpoints it may be only conditional (as $\sum(-1)^n/n$ ). The two endpoints need not agree.

$R$ is about the coefficients, not a chosen $x$ . The formula $1/R = \limsup|a_n|^{1/n}$ uses only the $a_n$ . Plugging a number for $x$ first and then "finding the radius" is a category error.

CounterexampleConvergence Need Not Be Uniform on the Whole Open Interval

Consider $\sum_{n=0}^\infty x^n = 1/(1-x)$ on $(-1,1)$ , with $R=1$ . One might hope the convergence is uniform on all of $(-1,1)$ . It is not: the remainder is $|x|^{N+1}/(1-x)$ , which is unbounded as $x\to 1^-$ for fixed $N$ , so no single $N$ works for all $x$ . The correct statement is that a power series converges uniformly on every closed subinterval $[c-r,c+r]$ with $r<R$ — not on the open interval itself. Uniform convergence holds on compact subsets strictly inside $R$ , which is exactly enough to justify term-by-term calculus, but mistaking "compact-subset uniform" for "uniform on $(c-R,c+R)$ " is a frequent error.

TheoremTerm-by-Term Differentiation and Integration

Let $f(x)=\sum_{n=0}^\infty a_n(x-c)^n$ have radius $R>0$ . Then $f$ is infinitely differentiable on $(c-R,c+R)$ , and

11

\quad f'(x) = \sum_{n=1}^\infty n\,a_n (x-c)^{n-1},

Derivation hint: differentiate term by term; the differentiated series has the same radius since $\lim n^{1/n}=1$ doesn't change the limsup.

12

\quad \int_c^x f(t)\,dt = \sum_{n=0}^\infty \frac{a_n}{n+1}(x-c)^{n+1},

Derivation hint: integrate term by term over a compact subinterval where convergence is uniform; the antiderivative series has the same radius $R$ . Both new series have the same radius of convergence $R$ . Furthermore the coefficients are recovered from the function by the Taylor formula $a_n = f^{(n)}(c)/n!$ .

connection

Power series sit at the crossroads of analysis. They are the engine of Taylor Series, which expand smooth functions as power series with coefficients $a_n=f^{(n)}(c)/n!$ ; not every smooth function equals its Taylor series, which leads to the subtler notion of Analytic Functions. The convergence theory here rests directly on Series Convergence Tests — the ratio, root, comparison, and alternating series tests applied to numerical series. In the complex setting, the interval of convergence becomes a disk and the subject blossoms into Complex Analysis, where every differentiable function is automatically analytic (a phenomenon with no real-variable analogue). Power series also provide the standard method for solving Ordinary Differential Equations near ordinary points, and the term-by-term calculus theorem ([11]–[12]) ties them to Uniform Convergence, the property that legitimizes interchanging limits, sums, derivatives, and integrals.

🔬 Where this appears

⏳ History

- 1668 — Nicolaus Mercator publishes the logarithmic series

\ln(1+x)=x-x^2/2+x^3/3-\cdots

, one of the first nontrivial power-series expansions in print.

- 1669–1671 — Isaac Newton, in De Analysi and Method of Fluxions, develops the generalized binomial series and treats power series as a universal computational tool, inverting and composing them freely.

- 1715 — Brook Taylor states Taylor's theorem, expressing functions via their derivatives at a point, though without convergence analysis.

- 1748 — Leonhard Euler's Introductio in analysin infinitorum uses power series with breathtaking (and sometimes reckless) freedom, deriving

e^{ix}=\cos x + i\sin x

and summing

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

.

- 1797–1813 — Joseph-Louis Lagrange attempts to base all of calculus on Taylor series (Théorie des fonctions analytiques); the program's gaps motivate later rigor.

- 1821 — Augustin-Louis Cauchy's Cours d'analyse introduces rigorous definitions of limit and convergence, the root and ratio tests, and the modern notion of radius of convergence.

- 1826 — Niels Henrik Abel proves Abel's Lemma and Abel's continuity theorem, putting the binomial series on firm footing and critiquing divergent-series misuse.

- 1892 — Jacques Hadamard publishes the Cauchy–Hadamard formula

1/R=\limsup|a_n|^{1/n}

in full generality, completing the determination of

R

.

- 1900s — Émile Borel, Alfred Tauber, and others develop summability methods (Abel, Cesàro, Borel summation) that assign meaningful values to certain divergent power series, extending the theory beyond classical convergence.

open-problem

The convergence theory described here is essentially complete for real and complex power series — but the boundary remains rich and partly mysterious.

Boundary behavior. What happens on the circle $|x-c|=R$ can be extraordinarily intricate. There exist power series convergent at every boundary point yet whose sum is nowhere analytically continuable past the circle (the circle is a natural boundary), e.g. $\sum z^{2^n}$ (a lacunary series, by Hadamard's gap theorem). Characterizing exactly which sets on the boundary are sets of convergence/divergence connects to deep results in Fourier analysis and harmonic analysis.

Summability of divergent series. A power series with $R=0$ — like Euler's $\sum n!\,x^n$ — diverges everywhere yet may be the asymptotic expansion of a genuine function (here related to the exponential integral). Borel summation and the theory of resurgence (Écalle) assign rigorous meaning to such series; understanding which divergent series are Borel-summable and how to resum them is an active area in mathematical physics (perturbation theory, quantum field theory).

Multivariate power series. For $\sum a_{\alpha}(x-c)^\alpha$ in several variables, "radius of convergence" is replaced by a domain of convergence that is logarithmically convex (a Reinhardt domain), and the geometry of convergence becomes a chapter of several complex variables with phenomena (Hartogs' extension) absent in one variable.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also