Geometric and Telescoping Series

How can we ever add up infinitely many numbers and get a finite answer? The question sounds paradoxical. But realize that the sum 11 is exactly 12+14+18+⋯\tfrac12 + \tfrac14 + \tfrac18 + \cdots. The ancient Greeks already puzzled over this in Zeno's paradox of Achilles and the tortoise. Geometric and telescoping series are the two families of infinite series where we can compute the partial sums in closed form. This is precisely what makes their convergence behavior transparent. It is provable from first principles, rather than merely plausible. These two families matter far beyond their own charm. They are the load-bearing walls of series theory. Every geometric series underlies the ratio and root tests. Every telescoping trick underlies integral estimates and the collapsing sums behind Taylor remainder formulas. Both appear constantly in probability (expected value of a geometric random variable), numerical analysis (error accumulation), and computer science (analysis of recursive algorithms). When a more complicated series is compared against a geometric or telescoping series, its fate is often decided instantly. This is the seed of the idea developed fully in The Comparison Test. The big picture is this: convergence of an infinite series ∑an\sum a_n is defined via the sequence of partial sums SN=∑n=1NanS_N = \sum_{n=1}^N a_n. This is established in the study of Infinite Series and Partial Sums. Most series resist any closed formula for SNS_N. This forces us to use inequalities and comparisons. Geometric and telescoping series are the exception. We can write SNS_N explicitly, take N→∞N \to \infty, and read off convergence or divergence with a single limit computation. This article gives the precise definitions of both series types. It proves the geometric series formula and its convergence criterion in full. It proves the telescoping series convergence criterion. It works two non-trivial numerical examples. It shows exactly what fails when the defining hypotheses are dropped.

Key takeaways
  • A geometric series ∑n=0∞arn\sum_{n=0}^\infty ar^n has partial sum SN=a1−rN1−rS_N = a\frac{1-r^N}{1-r} (for r≠1r \neq 1), converging to a1−r\frac{a}{1-r} iff ∣r∣<1|r| < 1.
  • A telescoping series ∑(bn−bn+1)\sum (b_n - b_{n+1}) has partial sum SN=b1−bN+1S_N = b_1 - b_{N+1}, converging iff lim⁡n→∞bn\lim_{n\to\infty} b_n exists.
  • Both series types are the rare case where the partial sum has a genuine closed form — no estimation needed.
  • Convergence of ∑arn\sum ar^n depends only on ∣r∣|r|; the sign and phase of rr affect only the limiting value, not the dichotomy.
  • These series are the benchmark against which the Comparison Test measures more complicated series.
  • Key idea: whenever partial sums collapse to a simple closed formula, convergence becomes a single limit computation instead of an estimation problem.

Where this lives today

Geometric series are the computational engine behind compound interest and annuities in finance: the present value of a perpetuity paying aa dollars per period at discount rate rr (with 0<r<10<r<1 per-period effective growth factor for the discounted payment) is exactly ∑n=0∞arn=a1−r\sum_{n=0}^\infty a r^n = \frac{a}{1-r}. In probability theory, the probability mass function of a geometric random variable (number of trials to first success with success probability pp) sums to 11 precisely because ∑n=1∞p(1−p)n−1=p1−(1−p)=1\sum_{n=1}^\infty p(1-p)^{n-1} = \frac{p}{1-(1-p)} = 1, and its expectation is computed via a differentiated geometric series. In computer science, the geometric series formula gives the total work of algorithms with geometrically decreasing subproblem sizes (e.g., analyzing recursive halving in divide-and-conquer algorithms, or the expected number of probes in certain hashing schemes). In numerical analysis, geometric decay bounds the error of iterative methods (e.g. fixed-point iteration with contraction constant r<1r<1 has error shrinking geometrically, guaranteeing convergence). Telescoping sums appear throughout calculus itself: the Fundamental Theorem of Calculus for Riemann sums, discrete analogues of integration by parts (Abel summation), and the proof that certain Taylor remainder terms vanish all rely on telescoping cancellation. In signal processing and control theory, geometric series describe the impulse response of stable linear time-invariant systems with exponentially decaying feedback. Both series types serve as the benchmark cases invoked constantly in the Comparison Test and its refinements (ratio test, root test), where a complicated series is bounded above by a geometric series to force convergence.

Ask the tutor: How does the geometric series formula generalize to power series with radius of convergence?

DefinitionGeometric series and telescoping series

Let a∈Ra \in \mathbb{R} (or C\mathbb{C}) and r∈Rr \in \mathbb{R} (or C\mathbb{C}). The geometric series with first term aa and common ratio rr is

∑n=0∞arn=a+ar+ar2+ar3+⋯ .\sum_{n=0}^{\infty} a r^n = a + ar + ar^2 + ar^3 + \cdots.

Its NN-th partial sum is SN=∑n=0N−1arnS_N = \sum_{n=0}^{N-1} ar^n (the sum of the first NN terms, indices 00 through N−1N-1). Let (bn)n=1∞(b_n)_{n=1}^\infty be a sequence of real (or complex) numbers. The telescoping series generated by (bn)(b_n) is

∑n=1∞(bn−bn+1).\sum_{n=1}^{\infty} (b_n - b_{n+1}).

Its NN-th partial sum is SN=∑n=1N(bn−bn+1)S_N = \sum_{n=1}^{N} (b_n - b_{n+1}). In both cases, as with any series, we say the series converges to SS if the sequence of partial sums (SN)(S_N) converges to SS as N→∞N \to \infty, and diverges otherwise.

Intuition

The geometric series is the algebra of self-similar shrinking (or growing). Each term is the previous term scaled by the same factor rr. If ∣r∣<1|r|<1, each new block of the sum is a fixed fraction of the remaining "gap" to the limit, so the partial sums approach a target the way a person walking half the remaining distance to a wall approaches the wall — never overshooting, always closing a fixed proportion of what's left. If ∣r∣≥1|r| \geq 1, terms fail to shrink relative to each other and the sum runs away or oscillates without settling. The telescoping series is a different mechanism entirely: it is not about repeated scaling but about cancellation. Writing the general term as a difference bn−bn+1b_n - b_{n+1} means that in the sum b1−b2+b2−b3+b3−b4+⋯b_1 - b_2 + b_2 - b_3 + b_3 - b_4 + \cdots, every middle term is added once and subtracted once — like a collapsible telescope, all the interior tubes vanish, leaving only the first piece b1b_1 and, in the limit, the "vanishing point" of the sequence bnb_n.

Figure
Geometric series: partial sums closing in on the limit a/(1−r)a/(1-r), here a=1,r=1/2a=1, r=1/2
S limit = 2 S₁=1 S₂=1.5 S₃=1.75 S₄=1.875 S₅=1.9375
Figure
Telescoping cancellation for bn−bn+1b_n - b_{n+1} with bn=1/nb_n = 1/n, showing interior terms cancel
Σ (1/n − 1/(n+1)) for n = 1..4 (1/1 − 1/2) + (1/2 − 1/3) + (1/3 − 1/4) + (1/4 − 1/5) cancelling pairs: −1/2+1/2, −1/3+1/3, −1/4+1/4 = 1/1 − 1/5 = 4/5 As N → ∞: 1/(N+1) → 0, so full sum → 1

Central theorems

TheoremGeometric Series Test

Let a,ra, r be real numbers with a≠0a \neq 0. Consider the series ∑n=0∞arn\sum_{n=0}^{\infty} ar^n. 1. If ∣r∣<1|r| < 1, the series converges and

∑n=0∞arn=a1−r.\sum_{n=0}^{\infty} ar^n = \frac{a}{1-r}.

2. If ∣r∣≥1|r| \geq 1, the series diverges.

Proof

Step 1: Closed form for the partial sum. Fix N≥1N \geq 1 and let

SN=∑n=0N−1arn=a+ar+ar2+⋯+arN−1.S_N = \sum_{n=0}^{N-1} ar^n = a + ar + ar^2 + \cdots + ar^{N-1}.

Case r=1r = 1: then SN=a⋅NS_N = a \cdot N, and we handle this case separately below. Case r≠1r \neq 1: multiply SNS_N by rr:

rSN=ar+ar2+⋯+arN−1+arN.rS_N = ar + ar^2 + \cdots + ar^{N-1} + ar^N.

Subtracting,

SN−rSN=a−arN,S_N - rS_N = a - ar^N,

since every term ar,ar2,…,arN−1ar, ar^2, \ldots, ar^{N-1} cancels between the two sums. Hence SN(1−r)=a(1−rN)S_N(1-r) = a(1-r^N), and since r≠1r \neq 1 we may divide by 1−r1-r to obtain the closed form

SN=a 1−rN1−r.(∗)S_N = a\,\frac{1-r^N}{1-r}. \tag{$\ast$}

This is an exact algebraic identity, valid for every N≥1N \geq 1 and every r≠1r \neq 1; no limiting process has been used yet.

Step 2: The case ∣r∣<1|r| < 1. We use the standard fact that for real or complex rr with ∣r∣<1|r| < 1, lim⁡N→∞rN=0\lim_{N\to\infty} r^N = 0. (Proof of this fact: given ε>0\varepsilon > 0, since ∣r∣<1|r| < 1 write ∣r∣=11+h|r| = \frac{1}{1+h} with h>0h > 0. By Bernoulli's inequality, (1+h)N≥1+Nh(1+h)^N \geq 1 + Nh, so

∣r∣N=1(1+h)N≤11+Nh→0|r|^N = \frac{1}{(1+h)^N} \leq \frac{1}{1+Nh} \to 0

as N→∞N \to \infty, since 11+Nh<ε\frac{1}{1+Nh} < \varepsilon once N>1h(1ε−1)N > \frac{1}{h}\left(\frac{1}{\varepsilon}-1\right). Hence ∣rN∣=∣r∣N→0|r^N| = |r|^N \to 0, so rN→0r^N \to 0.) Applying this to (∗)(\ast) and using continuity of the algebraic operations (limit of a constant times a convergent sequence, and limit of a constant minus a convergent sequence),

lim⁡N→∞SN=lim⁡N→∞a 1−rN1−r=a⋅1−01−r=a1−r.\lim_{N\to\infty} S_N = \lim_{N\to\infty} a\,\frac{1 - r^N}{1-r} = a\cdot\frac{1 - 0}{1-r} = \frac{a}{1-r}.

By definition, this means the series converges and its sum is a1−r\frac{a}{1-r}.

Step 3: The case r=1r = 1. Here SN=aNS_N = aN. Since a≠0a \neq 0, (SN)(S_N) is an unbounded (strictly monotonic in absolute value) sequence, hence does not converge to any finite limit. So the series diverges.

Step 4: The case r=−1r = -1. Here (∗)(\ast) gives, for even NN, SN=a1−12=0S_N = a\frac{1-1}{2} = 0, and for odd NN, SN=a1−(−1)2=aS_N = a\frac{1-(-1)}{2} = a. (Direct check: S1=aS_1 = a, S2=a−a=0S_2 = a - a = 0, S3=a−a+a=aS_3 = a - a + a = a, alternating.) Since a≠0a \neq 0, the subsequence of even-indexed partial sums converges to 00 while the subsequence of odd-indexed partial sums converges to a≠0a \neq 0. A convergent sequence has all its subsequences converging to the same limit, so (SN)(S_N) cannot converge. Hence the series diverges.

Step 5: The case ∣r∣>1|r| > 1. We show ∣SN∣→∞|S_N| \to \infty, so (SN)(S_N) cannot converge (a convergent sequence is bounded). From (∗)(\ast),

∣SN∣=∣a∣⋅∣1−rN∣∣1−r∣≥∣a∣⋅∣r∣N−1∣1−r∣|S_N| = |a| \cdot \frac{|1 - r^N|}{|1-r|} \geq |a|\cdot \frac{|r|^N - 1}{|1-r|}

using the reverse triangle inequality ∣1−rN∣≥∣rN∣−1=∣r∣N−1|1 - r^N| \geq |r^N| - 1 = |r|^N - 1. Since ∣r∣>1|r| > 1, we have ∣r∣N→∞|r|^N \to \infty (this follows from Bernoulli's inequality: writing ∣r∣=1+h|r| = 1+h with h>0h>0, ∣r∣N≥1+Nh→∞|r|^N \geq 1+Nh \to \infty). Hence the lower bound ∣a∣∣r∣N−1∣1−r∣→∞|a|\frac{|r|^N-1}{|1-r|} \to \infty, forcing ∣SN∣→∞|S_N| \to \infty. A sequence tending to infinity in absolute value is unbounded, hence not convergent. So the series diverges. Steps 3–5 together cover exactly the cases ∣r∣≥1|r| \geq 1 (namely r=1r=1, r=−1r=-1, and ∣r∣>1|r|>1 inclusive of complex rr off the unit circle in magnitude, real r<−1r < -1, and real r>1r > 1), completing the proof. ■\blacksquare

∎
TheoremTelescoping Series Convergence Criterion

Let (bn)n=1∞(b_n)_{n=1}^\infty be a sequence of real numbers. The telescoping series ∑n=1∞(bn−bn+1)\sum_{n=1}^\infty (b_n - b_{n+1}) converges if and only if (bn)(b_n) converges, in which case

∑n=1∞(bn−bn+1)=b1−lim⁡n→∞bn.\sum_{n=1}^{\infty} (b_n - b_{n+1}) = b_1 - \lim_{n\to\infty} b_n.
Proof

Step 1: Closed form for the partial sum, by induction. Let SN=∑n=1N(bn−bn+1)S_N = \sum_{n=1}^{N} (b_n - b_{n+1}). We claim

SN=b1−bN+1for every N≥1.(∗∗)S_N = b_1 - b_{N+1} \quad \text{for every } N \geq 1. \tag{$\ast\ast$}

Base case N=1N=1: S1=b1−b2S_1 = b_1 - b_2, which matches (∗∗)(\ast\ast) exactly. Inductive step: suppose (∗∗)(\ast\ast) holds for some N=kN = k, i.e. Sk=b1−bk+1S_k = b_1 - b_{k+1}. Then

Sk+1=Sk+(bk+1−bk+2)=(b1−bk+1)+(bk+1−bk+2)=b1−bk+2,S_{k+1} = S_k + (b_{k+1} - b_{k+2}) = (b_1 - b_{k+1}) + (b_{k+1} - b_{k+2}) = b_1 - b_{k+2},

which is exactly (∗∗)(\ast\ast) with N=k+1N = k+1. By the principle of mathematical induction, (∗∗)(\ast\ast) holds for all N≥1N \geq 1.

Step 2: (⇐\Leftarrow) If (bn)(b_n) converges, the series converges. Suppose lim⁡n→∞bn=L\lim_{n\to\infty} b_n = L exists (a finite real number). Then, since the index N+1→∞N+1 \to \infty as N→∞N \to \infty, the subsequence (bN+1)N=1∞(b_{N+1})_{N=1}^\infty also converges to LL (a sequence and any of its tail-subsequences share the same limit). By (∗∗)(\ast\ast) and the algebra of limits,

lim⁡N→∞SN=lim⁡N→∞(b1−bN+1)=b1−L.\lim_{N\to\infty} S_N = \lim_{N\to\infty} (b_1 - b_{N+1}) = b_1 - L.

So (SN)(S_N) converges, meaning by definition the series converges, and its value is b1−Lb_1 - L.

Step 3: (⇒\Rightarrow) If the series converges, (bn)(b_n) converges. Suppose (SN)(S_N) converges to some limit SS. By (∗∗)(\ast\ast), bN+1=b1−SNb_{N+1} = b_1 - S_N for every N≥1N \geq 1. Since (SN)(S_N) converges to SS, the sequence (b1−SN)N=1∞(b_1 - S_N)_{N=1}^\infty converges to b1−Sb_1 - S (again by the algebra of limits). Hence (bN+1)N=1∞(b_{N+1})_{N=1}^\infty converges to b1−Sb_1 - S. But (bN+1)N=1∞(b_{N+1})_{N=1}^\infty is precisely the tail of the sequence (bn)n=1∞(b_n)_{n=1}^\infty starting at n=2n=2; a sequence converges if and only if one of its tails converges (to the same limit), so (bn)(b_n) itself converges, and

lim⁡n→∞bn=b1−S.\lim_{n\to\infty} b_n = b_1 - S.

Step 4: Conclusion. Combining Steps 2 and 3, the series converges if and only if (bn)(b_n) converges. When it does, Step 2 gives the sum as b1−Lb_1 - L where L=lim⁡n→∞bnL = \lim_{n\to\infty} b_n, matching the claimed formula. ■\blacksquare

∎
CorollaryDivergence criterion for telescoping series

If (bn)(b_n) diverges (fails to converge to any finite real limit), then ∑n=1∞(bn−bn+1)\sum_{n=1}^\infty (b_n - b_{n+1}) diverges.

This is immediate from the "only if" direction of the theorem (Step 3), by contraposition: if the series converged, (bn)(b_n) would converge; so if (bn)(b_n) diverges, the series must diverge.

Worked examples

Example 1 (Geometric series with a non-trivial ratio and shifted start). Evaluate

∑n=2∞3⋅(−2)n5n+1.\sum_{n=2}^{\infty} \frac{3\cdot(-2)^n}{5^{n+1}}.

First put this in the standard form. Write the general term as

3⋅(−2)n5n+1=35⋅(−2)n5n=35(−25)n.\frac{3\cdot(-2)^n}{5^{n+1}} = \frac{3}{5}\cdot\frac{(-2)^n}{5^n} = \frac{3}{5}\left(\frac{-2}{5}\right)^n.

So the series is ∑n=2∞c(−25)n\sum_{n=2}^{\infty} c \left(\frac{-2}{5}\right)^n with c=35c = \frac{3}{5}, but the sum starts at n=2n=2 rather than n=0n=0. Reindex with k=n−2k = n-2, so n=k+2n = k+2 and kk runs from 00 to ∞\infty:

∑n=2∞35(−25)n=∑k=0∞35(−25)k+2=∑k=0∞35(−25)2⏟=a(−25)k.\sum_{n=2}^{\infty} \frac{3}{5}\left(\frac{-2}{5}\right)^n = \sum_{k=0}^{\infty} \frac{3}{5}\left(\frac{-2}{5}\right)^{k+2} = \sum_{k=0}^{\infty} \underbrace{\frac{3}{5}\left(\frac{-2}{5}\right)^2}_{=a} \left(\frac{-2}{5}\right)^{k}.

Compute aa: (−25)2=425\left(\frac{-2}{5}\right)^2 = \frac{4}{25}, so a=35⋅425=12125a = \frac{3}{5}\cdot\frac{4}{25} = \frac{12}{125}. Here r=−25r = -\frac{2}{5}, and ∣r∣=25<1|r| = \frac25 < 1, so by the Geometric Series Test (Step 2 of the proof above) this converges to

a1−r=12/1251−(−2/5)=12/1257/5=12125⋅57=60875=12175.\frac{a}{1-r} = \frac{12/125}{1-(-2/5)} = \frac{12/125}{7/5} = \frac{12}{125}\cdot\frac{5}{7} = \frac{60}{875} = \frac{12}{175}.

Check numerically. The first few terms (at n=2,3,4n=2,3,4) are

3(−2)253=12125=0.0960,3(−2)354=−24625=−0.0384,3(−2)455=483125=0.01536.\frac{3(-2)^2}{5^3} = \frac{12}{125} = 0.0960, \quad \frac{3(-2)^3}{5^4} = \frac{-24}{625} = -0.0384, \quad \frac{3(-2)^4}{5^5} = \frac{48}{3125} = 0.01536.

Partial sum through n=4n=4: 0.0960−0.0384+0.01536=0.072960.0960 - 0.0384 + 0.01536 = 0.07296. Compare to the exact value 12175≈0.068571\frac{12}{175} \approx 0.068571. The partial sums are oscillating and converging toward 0.06860.0686, consistent with r<0r<0 (alternating overshoot) and ∣r∣=0.4|r| = 0.4 (moderately fast convergence) — matching the closed form.

Example 2 (Telescoping series requiring partial fractions). Evaluate

∑n=1∞5(3n−2)(3n+1).\sum_{n=1}^{\infty} \frac{5}{(3n-2)(3n+1)}.

Step 1: Decompose into a telescoping form. We seek constants A,BA, B with

5(3n−2)(3n+1)=A3n−2−A3n+1\frac{5}{(3n-2)(3n+1)} = \frac{A}{3n-2} - \frac{A}{3n+1}

matching the shape bn−bn+1b_n - b_{n+1} with bn=A3n−2b_n = \frac{A}{3n-2} (note bn+1=A3(n+1)−2=A3n+1b_{n+1} = \frac{A}{3(n+1)-2} = \frac{A}{3n+1}, exactly the second term — this confirms the shift is consistent). Combine the right side over a common denominator:

A3n−2−A3n+1=A(3n+1)−A(3n−2)(3n−2)(3n+1)=3A(3n−2)(3n+1).\frac{A}{3n-2} - \frac{A}{3n+1} = \frac{A(3n+1) - A(3n-2)}{(3n-2)(3n+1)} = \frac{3A}{(3n-2)(3n+1)}.

Matching numerators, 3A=53A = 5, so A=53A = \frac{5}{3}. Thus

5(3n−2)(3n+1)=5/33n−2−5/33n+1=bn−bn+1,bn:=53(3n−2).\frac{5}{(3n-2)(3n+1)} = \frac{5/3}{3n-2} - \frac{5/3}{3n+1} = b_n - b_{n+1}, \quad b_n := \frac{5}{3(3n-2)}.

Step 2: Apply the telescoping formula. By (∗∗)(\ast\ast) in the proof above,

SN=∑n=1N(bn−bn+1)=b1−bN+1=53(3⋅1−2)−53(3(N+1)−2)=53−53(3N+1).S_N = \sum_{n=1}^{N} (b_n - b_{n+1}) = b_1 - b_{N+1} = \frac{5}{3(3\cdot1-2)} - \frac{5}{3(3(N+1)-2)} = \frac{5}{3} - \frac{5}{3(3N+1)}.

As a numeric check take N=3N = 3 directly from the original series:

51⋅4+54⋅7+57⋅10=1.25+0.178571+0.071429=1.5.\frac{5}{1\cdot 4} + \frac{5}{4\cdot 7} + \frac{5}{7\cdot 10} = 1.25 + 0.178571 + 0.071429 = 1.5.

From the formula, S3=53−53(10)=1.666667−0.166667=1.5S_3 = \frac53 - \frac{5}{3(10)} = 1.666667 - 0.166667 = 1.5. The two agree exactly.

Step 3: Take the limit. Since bN+1=53(3N+1)→0b_{N+1} = \frac{5}{3(3N+1)} \to 0 as N→∞N \to \infty (the denominator grows without bound), by the Telescoping Series Convergence Criterion the series converges, with sum

lim⁡N→∞SN=53−0=53.\lim_{N\to\infty} S_N = \frac{5}{3} - 0 = \frac{5}{3}.

Counterexample

CounterexampleDropping "(bn)(b_n) converges" for a telescoping series: bounded is not enough

The convergence criterion for telescoping series requires the full sequence (bn)(b_n) to converge to a single limit LL — it is not enough for (bn)(b_n) to merely be bounded, or to have convergent subsequences with different limits. Consider

bn=(−1)n.b_n = (-1)^n.

This sequence is bounded (∣bn∣≤1|b_n| \le 1 for all nn) but does not converge, since it oscillates between 11 and −1-1 forever, and any convergent sequence has a unique limit shared by all its subsequences — here the even-indexed terms tend to 11 while odd-indexed terms tend to −1-1. Form the telescoping series ∑n=1∞(bn−bn+1)\sum_{n=1}^\infty (b_n - b_{n+1}). Its terms are

b1−b2=−1−1=−2,b2−b3=1−(−1)=2,b3−b4=−1−1=−2,…b_1 - b_2 = -1-1 = -2,\quad b_2-b_3 = 1-(-1)=2,\quad b_3-b_4=-1-1=-2,\quad \ldots

so the series is −2+2−2+2−⋯-2 + 2 - 2 + 2 - \cdots. By the closed form (∗∗)(\ast\ast) proven above, SN=b1−bN+1=−1−bN+1S_N = b_1 - b_{N+1} = -1 - b_{N+1}, which equals −1−1=−2-1-1=-2 when N+1N+1 is even and −1−(−1)=0-1-(-1)=0 when N+1N+1 is odd. So SNS_N alternates forever between −2-2 and 00, never settling — the series diverges, exactly as the theorem predicts, since (bn)(b_n) fails to converge. This shows the hypothesis "(bn)(b_n) converges" cannot be weakened to "(bn)(b_n) bounded": boundedness alone gives a bounded but oscillating, non-convergent, sequence of partial sums.

Scope, edge cases & common traps

SituationWhat happensWhy
∥r∥<1\|r\| < 1, geometricConverges to a/(1−r)a/(1-r)Step 2 of geometric proof; rN→0r^N \to 0
r=1r = 1Diverges (to ±∞\pm\infty if a≠0a \ne 0)SN=aNS_N = aN unbounded
r=−1r = -1Diverges by oscillationEven/odd subsequences differ
∥r∥>1\|r\| > 1Diverges (unbounded)∥r∥N→∞\|r\|^N \to \infty
a=0a = 0Trivially converges to 00 for any rrEvery term is 00
Telescoping, (bn)(b_n) convergentConverges to b1−lim⁡bnb_1 - \lim b_nMain theorem
Telescoping, (bn)(b_n) divergent (e.g. oscillating or →∞\to\infty)DivergesCorollary, contrapositive
Telescoping with a different offset, e.g. bn−bn+2b_n - b_{n+2}Not directly covered — must re-derive partial sum (it will telescope in pairs, leaving two head and two tail terms)Formula (∗∗)(\ast\ast) assumes consecutive-index differences

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered
AC
Augustin-Louis Cauchy
1789–1857 · Founder of rigorous analysis in 19th-century France

Cauchy was one of the principal architects of the rigorous ε\varepsilon–NN framework that makes statements like "the geometric series converges iff ∣r∣<1|r|<1" a genuine theorem rather than a computational habit. Before Cauchy's Cours d'Analyse (1821), mathematicians manipulated infinite series with real virtuosity but with no agreed-upon definition of "sum of an infinite series" — Euler famously derived (correct, in a different formal sense) results from divergent series such as 1−1+1−1+⋯1 - 1 + 1 - 1 + \cdots. Cauchy insisted that a series's sum is, by definition, the limit of its partial sums, and that this limit must be shown to exist via precise inequalities. This is exactly the definitional backbone used throughout this article: the geometric series formula and the telescoping criterion are both proved by explicit partial-sum formulas fed into a rigorous limit, in the Cauchy tradition. The geometric series itself, however, is far older — its partial-sum identity appears already in Euclid's Elements (Book IX, Proposition 35), and the paradoxes of infinite summation trace back to Zeno of Elea around 450 BCE.

History
- c. 450 BCE — Zeno of Elea poses the dichotomy paradox, implicitly invoking ∑12n\sum \frac12{}^{n}.
- c. 300 BCE — Euclid's Elements, Book IX, gives the geometric partial-sum identity in essentially the algebraic form used above.
- 1350s — Nicole Oresme proves the harmonic series diverges and studies geometric-like series with medieval rigor.
- 1600s — Newton, Leibniz, and contemporaries use geometric series freely to represent functions (precursors to power series).
- 1821 — Cauchy's Cours d'Analyse defines series convergence via partial sums with full epsilon-style rigor, the modern definition used here.
Why does the geometric series need ∣r∣<1|r| < 1 and not just r≠1r \neq 1?

Because divergence also occurs at r=−1r = -1 and at ∣r∣>1|r| > 1, not only at r=1r = 1. At r=1r=1 the partial sums grow linearly to infinity; at r=−1r=-1 they oscillate between two distinct values without settling; and for ∣r∣>1|r|>1 they blow up in absolute value, oscillating in sign if rr is negative. The theorem's proof handles all three failure modes separately (Steps 3, 4, 5 above), and each genuinely needs a different argument: unboundedness, subsequence-limit mismatch, and unboundedness via Bernoulli's inequality respectively. The single clean condition ∣r∣<1|r|<1 is exactly the boundary that rules out every one of these failure modes simultaneously, since ∣r∣<1|r|<1 is equivalent to rN→0r^N \to 0.

Does every convergent series have some telescoping representation?

In a formal sense, yes: given any convergent series ∑an\sum a_n with partial sums SN→SS_N \to S, define bn=S−Sn−1b_n = S - S_{n-1} (with S0=0S_0 = 0); one can check bn−bn+1=Sn−Sn−1=anb_n - b_{n+1} = S_{n} - S_{n-1} = a_n, so the series is "telescoping" for this particular choice of bnb_n. But this is a tautology, not a useful tool — the whole point of recognizing a series as telescoping is that you can find a bnb_n with a genuinely simpler closed form than the partial sums themselves, typically via partial fraction decomposition or algebraic manipulation, as in Example 2. Without an explicit simple bnb_n, "telescoping" carries no computational advantage.

What is the relationship between geometric series and the ratio test for general series?

The ratio test examines lim⁡n→∞∣an+1an∣=L\lim_{n\to\infty} \left|\frac{a_{n+1}}{a_n}\right| = L and concludes convergence if L<1L<1, divergence if L>1L>1. The proof works by comparing the tail of ∑an\sum a_n to a genuine geometric series with ratio close to LL: once ∣an+1∣∣an∣\frac{|a_{n+1}|}{|a_n|} stays below some r<1r<1 for all large nn, the terms are eventually bounded by ∣aN∣rn−N|a_N| r^{n-N}, a geometric series, which converges. So the geometric series is not just an example but the literal comparison object baked into the ratio test's proof — a preview of the general machinery in the Comparison Test.

Can a telescoping series have bn→∞b_n \to \infty and still converge?

No. If (bn)(b_n) diverges to +∞+\infty (or −∞-\infty), it does not converge to a finite limit, so by the "only if" direction of the theorem (Step 3 of the proof), the series ∑(bn−bn+1)\sum (b_n - b_{n+1}) diverges. For example, bn=nb_n = n gives bn−bn+1=−1b_n - b_{n+1} = -1 for every nn, so the series is −1−1−1−⋯-1-1-1-\cdots, whose partial sums SN=−N→−∞S_N = -N \to -\infty; indeed (∗∗)(\ast\ast) gives SN=b1−bN+1=1−(N+1)=−NS_N = b_1 - b_{N+1} = 1-(N+1) = -N, matching directly and diverging as predicted.

Why does reindexing (shifting the starting index) not change whether a geometric series converges?

Convergence is a statement about the tail behavior of partial sums, and shifting the start index only rescales the first term aa while leaving the ratio rr unchanged — as Example 1 shows explicitly, ∑n=2∞crn\sum_{n=2}^\infty c r^n is again a geometric series with a new (computable) first term a=cr2a = cr^2 and the same ratio rr. Since the convergence dichotomy in the Geometric Series Test depends only on ∣r∣|r|, not on aa, reindexing changes the numeric value of the sum but never changes whether it converges. This is a special case of the general fact that convergence of a series is unaffected by discarding or adding finitely many terms.

Is it true that if ∑an\sum a_n converges then ana_n must be expressible as bn−bn+1b_n - b_{n+1} for some nice (bn)(b_n)?

Not in general with a "nice" closed form — the tautological construction from the second FAQ always works formally, but producing an explicit, simple bnb_n (as required to make the telescoping method useful) is only possible when the terms ana_n have special algebraic structure, most commonly when ana_n is a rational function of nn that admits a partial fraction decomposition into consecutive reciprocal-type terms, as in Example 2. Most convergent series (e.g. ∑1/n2\sum 1/n^2) have no elementary telescoping form, which is exactly why other tools such as comparison, integral, and ratio tests are needed for the majority of series encountered in practice.

What's the fastest way to recognize whether a given series is geometric or telescoping?

For geometric: compute the ratio of consecutive terms an+1/ana_{n+1}/a_n; if it is a constant independent of nn, the series is geometric with that constant as rr. For telescoping: try to write the general term ana_n as a difference f(n)−f(n+1)f(n) - f(n+1) for some explicit function ff; a strong signal that this is possible is when ana_n is a rational function whose denominator factors into linear terms differing by a fixed integer step (like (3n−2)(3n+1)(3n-2)(3n+1) in Example 2), since partial fractions then naturally produce the telescoping shape. If neither pattern applies, the series is neither, and other convergence tests (comparison, ratio, root, integral) become necessary.

Before this

Where this leads