An infinite series is not an instruction to “add forever”. It is a sequence of finite partial sums whose convergence must be proved.
The preceding R25 guides built limits, derivatives and integration from finite approximations plus controlled limiting processes. Series use the same architecture in a discrete setting. We first define partial sums. Only then do convergence tests tell us whether those partial sums approach a finite limit.
This guide therefore separates three questions that are often collapsed: do the individual terms tend to zero, do the partial sums converge, and if they converge, how safely may the sum be manipulated, rearranged, differentiated or integrated?
Level: university-entry / undergraduate real-analysis enrichment. Prerequisites: sequences and limits from R25.01, basic integration from R25.03, and elementary algebra.
Reading route: partial sums → necessary term test → geometric and telescoping series → harmonic and p-series → Cauchy criterion → comparison → integral test → ratio/root tests → absolute convergence → alternating series → rearrangements → products → power series → radius and endpoints → termwise calculus → Taylor series → theorem boundaries → practice → solutions.
1. Infinite sums are defined through partial sums
Given a sequence (a_n), define the N-th partial sum
S_N=Σ_{n=1}^N a_n.
The infinite series Σ_{n=1}^∞a_n converges to S if the sequence of partial sums S_N converges to S.
If the partial sums fail to approach a finite limit, the series diverges.
Every rigorous statement about an infinite series therefore reduces to a statement about an ordinary sequence.
2. The nth-term test is necessary but not sufficient
If Σa_n converges, then a_n→0.
Indeed a_n=S_n−S_{n−1}; if both partial sums tend to the same S, their difference tends to zero.
Therefore if a_n does not tend to zero, the series certainly diverges.
But a_n→0 does not prove convergence. The harmonic series Σ1/n is the standard counterexample.
3. Geometric series give the first exact convergence model
For r≠1, the finite geometric sum is
1+r+⋯+r^N=(1−r^{N+1})/(1−r).
If |r|<1, then r^{N+1}→0, so
Σ_{n=0}^∞r^n=1/(1−r).
If |r|≥1, the terms fail to tend to zero or the partial sums fail to converge. Thus the exact convergence condition is |r|<1.
4. The tail of a geometric series gives an exact error formula
For |r|<1, after summing through n=N, the remaining tail is
R_N=Σ_{n=N+1}^∞r^n=r^{N+1}/(1−r).
Taking absolute values gives a certified truncation error.
This is the discrete analogue of a Taylor remainder or numerical error bound: approximation becomes useful when its omitted tail is controlled.
5. Telescoping series converge because cancellation is visible in partial sums
Consider
Σ_{n=1}^∞1/[n(n+1)].
Partial fractions give 1/[n(n+1)]=1/n−1/(n+1).
Therefore
S_N=(1−1/2)+(1/2−1/3)+⋯+(1/N−1/(N+1))=1−1/(N+1).
Hence S_N→1. The series converges to 1.
6. The harmonic series diverges despite terms tending to zero
The harmonic series is Σ1/n.
Group terms by powers of two:
1 +(1/2) +(1/3+1/4) +(1/5+⋯+1/8)+⋯
Each block after the first contains 2^{k−1} terms, each at least 1/2^k, so each block contributes at least 1/2.
Partial sums therefore grow without bound. The terms vanish, but too slowly.
7. p-series have a sharp convergence threshold
The series
Σ_{n=1}^∞1/n^p
converges exactly when p>1.
For p≤0, the terms do not even tend to zero. For 0<p≤1, comparison or the integral test shows divergence. For p>1, the integral test gives convergence.
This discrete threshold mirrors the improper integral ∫₁^∞x^{-p}dx from R25.03.
8. The Cauchy criterion tests tails without knowing the final sum
A series Σa_n converges in R exactly when for every ε>0 there exists N such that m>n≥N implies
|a_{n+1}+a_{n+2}+⋯+a_m|<ε.
This is simply the Cauchy criterion applied to partial sums.
It expresses convergence as a tail-control property: sufficiently late finite chunks contribute arbitrarily little net change.
9. Direct comparison transfers convergence for nonnegative series
Suppose 0≤a_n≤b_n eventually.
If Σb_n converges, then Σa_n converges.
If Σa_n diverges and a_n≤b_n, then Σb_n diverges.
The order direction must be matched to the conclusion. A smaller divergent benchmark proves nothing about a still smaller series; a larger convergent benchmark proves nothing about a still larger one.
10. Worked comparison: Σ1/(n²+3)
For n≥1,
0<1/(n²+3)≤1/n².
Because Σ1/n² converges, comparison proves Σ1/(n²+3) converges.
The comparison proves convergence without needing the exact sum.
11. Limit comparison identifies shared asymptotic scale
For positive a_n,b_n, if
lim_{n→∞}a_n/b_n=c
with 0<c<∞, then Σa_n and Σb_n have the same convergence behaviour.
Example: a_n=(3n+1)/(n³+2). Then a_n/(1/n²)→3, so Σa_n converges like the p-series Σ1/n².
12. The integral test connects discrete and continuous tails
If f is positive, continuous and decreasing for x≥N and a_n=f(n), then
Σ_{n=N}^∞a_n and ∫_N^∞f(x)dx either both converge or both diverge.
The proof compares rectangles to the area under a decreasing graph.
The monotonicity and positivity hypotheses are part of the theorem; the integral test is not a generic replacement for every series.
13. Integral-test remainder bounds estimate the unsummed tail
Under the integral-test hypotheses, if R_N=Σ_{n=N+1}^∞a_n, then
∫_{N+1}^∞f(x)dx≤R_N≤∫_N^∞f(x)dx.
Thus an integral does more than decide convergence; it can bound the truncation error after finitely many terms.
14. The ratio test detects exponential-scale decay or growth
For a series Σa_n, suppose
L=lim_{n→∞}|a_{n+1}/a_n|
exists.
- If L<1, the series converges absolutely.
- If L>1, or the ratio is eventually greater than one in the appropriate sense, the terms fail to decay sufficiently and the series diverges.
- If L=1, the test is inconclusive.
Both Σ1/n and Σ1/n² have ratio limit 1, yet one diverges and the other converges. The boundary really is inconclusive.
15. The root test is the nth-root analogue
Let
L=limsup_{n→∞}|a_n|^{1/n}.
- If L<1, Σa_n converges absolutely.
- If L>1, the series diverges.
- If L=1, the test is inconclusive.
The limsup formulation handles cases where ordinary nth-root limits fail to exist.
16. Absolute convergence makes signs irrelevant to convergence
A series Σa_n converges absolutely if Σ|a_n| converges.
Absolute convergence implies ordinary convergence because tails satisfy
|Σa_n|≤Σ|a_n|.
Absolute convergence is stronger than convergence and gives much safer algebraic manipulation properties.
17. Conditional convergence depends on cancellation
A series converges conditionally if Σa_n converges but Σ|a_n| diverges.
The alternating harmonic series
1−1/2+1/3−1/4+⋯
is the canonical example. Its absolute-value series is harmonic and diverges, while its alternating partial sums converge.
18. The alternating-series test uses decreasing error envelopes
If b_n≥0, b_{n+1}≤b_n eventually and b_n→0, then
Σ(−1)^{n−1}b_n
converges.
Moreover, if S_N is the N-th partial sum and S the limit, then
|S−S_N|≤b_{N+1}.
The first omitted term controls the truncation error.
19. Rearranging absolutely convergent series preserves the sum
If Σa_n converges absolutely, any rearrangement of its terms converges to the same sum.
Absolute tail control prevents a permutation from changing the limiting value.
This is one reason absolute convergence is a structural safety condition rather than merely a stronger numerical test.
20. Conditional series can change under rearrangement
Riemann’s rearrangement theorem says a conditionally convergent real series can be rearranged to converge to any prescribed real number, or rearranged to diverge in various ways.
The phenomenon occurs because the positive and negative subseries each diverge, while their original order creates delicate cancellation.
Therefore the informal rule “addition is commutative, so order never matters” fails for infinite conditionally convergent sums.
21. The Cauchy product needs convergence hypotheses
For series Σa_n and Σb_n, the Cauchy product has coefficients
c_n=Σ_{k=0}^n a_kb_{n−k}.
If both series converge absolutely, then the Cauchy product converges absolutely and its sum is the product of the two sums.
Weaker results exist—for example, convergence of both series with one absolutely convergent is sufficient—but unrestricted multiplication of arbitrary convergent series is not justified.
22. Power series are series of functions
A power series centred at a is
Σ_{n=0}^∞c_n(x−a)^n.
For each fixed x this becomes an ordinary numerical series.
The central question is now which x-values make that numerical series converge.
23. Every power series has a radius of convergence
There exists R∈[0,∞] such that the power series converges absolutely for |x−a|<R and diverges for |x−a|>R.
The endpoints |x−a|=R are not decided by the radius theorem and must be tested separately.
The Cauchy–Hadamard formula gives
1/R=limsup_{n→∞}|c_n|^{1/n}
with the standard conventions for zero and infinity.
24. Ratio tests often compute the radius quickly
For Σc_n(x−a)^n, applying the ratio test can produce a condition of the form |x−a|<R.
Example:
Σ x^n/n has ratio magnitude tending to |x|, so the radius is R=1.
At x=1 the series becomes harmonic and diverges. At x=−1 it becomes alternating harmonic and converges conditionally.
The interval of convergence is therefore [−1,1), with different endpoint behaviours.
25. Power series are exceptionally well behaved inside their radius
Inside |x−a|<R, a power series may be differentiated and integrated term by term.
The differentiated and integrated power series have the same radius of convergence R.
Thus if
f(x)=Σc_n(x−a)^n,
then inside the radius
f'(x)=Σ_{n=1}^∞nc_n(x−a)^{n−1}.
The endpoint behaviour still needs separate analysis.
26. Uniform convergence explains why some limit operations are safe
Pointwise convergence controls each x separately. Uniform convergence requires one index threshold to work across the whole domain.
Uniform convergence is strong enough to preserve continuity under uniform limits of continuous functions and supports interchange of limit and integral under standard hypotheses.
This guide uses uniform convergence only as a bridge. R25.08 owns the deeper question of when limits, derivatives, integrals and infinite sums may be exchanged.
27. The Weierstrass M-test gives a practical uniform-convergence criterion
If |f_n(x)|≤M_n for every x in a domain and ΣM_n converges, then Σf_n converges uniformly and absolutely.
The theorem converts a function-series problem into an ordinary positive numerical series.
It is especially useful on compact subintervals strictly inside a power series’ radius of convergence.
28. Taylor series are power series built from derivatives
For a smooth function f, its formal Taylor series at a is
Σ_{n=0}^∞f^{(n)}(a)(x−a)^n/n!.
Taylor’s theorem gives finite polynomials plus remainders. Equality between f and the infinite Taylor series requires the remainder to tend to zero on the region of interest.
As R25.02 emphasised, infinitely differentiable does not automatically mean analytic.
29. Worked exponential series
For e^x, all derivatives at zero equal 1, so the Taylor series is
Σ_{n=0}^∞x^n/n!.
The ratio of consecutive absolute terms is |x|/(n+1)→0 for every fixed x.
Therefore the series converges absolutely for every x: radius R=∞.
Taylor remainder estimates show the series actually equals e^x everywhere.
30. A convergence test is not a sum-finding method
Comparison, ratio, root and integral tests usually decide whether a sum exists; they do not generally produce its exact value.
Geometric and telescoping structures are special because their partial sums can often be simplified explicitly.
Keeping “existence of a finite sum” separate from “closed-form evaluation” prevents an important category error.
31. A dependable convergence-test workflow
First check whether a_n→0. If not, stop: the series diverges.
Then inspect structure: geometric, telescoping, p-series, alternating, factorial/exponential, rational-in-n, or a power series.
For positive terms, comparison and integral methods are often natural. For factorials or exponentials, ratio/root tests are often efficient.
If signs alternate, ask whether absolute convergence holds before using a conditional test.
For power series, determine the radius first and test boundary points individually.
If a test returns its boundary case—especially ratio/root value 1—switch tests rather than forcing a conclusion.
32. Common misconceptions
a_n→0 proves Σa_n converges. False: harmonic series.
Ratio-test limit 1 means divergence. False; it means the test is inconclusive.
Every convergent series can be freely rearranged. False for conditional convergence.
The radius of a power series decides the endpoints. No; endpoints require separate tests.
Every smooth function equals its Taylor series. False.
A convergence test gives the exact sum. Usually it only establishes convergence or divergence.
33. Independent practice: twenty questions
- Define convergence of Σa_n using partial sums.
- Why must a_n→0 if Σa_n converges?
- For what r does Σr^n converge?
- Find Σ_{n=0}^∞(1/3)^n.
- Evaluate Σ_{n=1}^∞1/[n(n+1)].
- Explain one proof that the harmonic series diverges.
- For what p does Σ1/n^p converge?
- State the Cauchy criterion for a series.
- Use comparison to decide Σ1/(n²+3).
- State the limit-comparison test.
- State the integral test hypotheses.
- What does ratio-test limit L<1 imply?
- What does ratio-test limit L=1 imply?
- Define absolute convergence.
- Give a conditionally convergent series.
- State the alternating-series error estimate.
- What can rearrangement do to a conditionally convergent real series?
- Define the radius of convergence of a power series.
- Why must power-series endpoints be checked separately?
- What is the radius of Σx^n/n!?
34. Worked solutions and checks
1. Let S_N=Σ_{n=1}^N a_n. The series converges to S when S_N→S.
2. a_n=S_n−S_{n−1}, and both partial sums tend to the same limit.
3. |r|<1.
4. 1/(1−1/3)=3/2.
5. It telescopes to 1.
6. Group terms from 2^{k−1}+1 through 2^k; each block contributes at least about 1/2, so partial sums are unbounded.
7. Exactly when p>1.
8. Every sufficiently late finite tail must have absolute sum smaller than every prescribed ε.
9. 0<1/(n²+3)≤1/n² and Σ1/n² converges, so the given series converges.
10. For positive terms, if a_n/b_n→c with 0<c<∞, the two series have the same convergence behaviour.
11. f positive, continuous and decreasing eventually, with a_n=f(n).
12. Absolute convergence.
13. Nothing decisive; use another test.
14. Σ|a_n| converges.
15. The alternating harmonic series.
16. For decreasing b_n→0, the error after N terms is at most b_{N+1}.
17. By Riemann’s rearrangement theorem it can be rearranged to many different real sums or made to diverge.
18. R is the number such that the series converges absolutely for |x−a|<R and diverges for |x−a|>R.
19. The radius theorem is silent at |x−a|=R; different endpoints can have different behaviour.
20. R=∞ because |a_{n+1}/a_n|=|x|/(n+1)→0 for every fixed x.
35. Batch 01 handoff
The first four R25 cells now form one rigorous route: limits and continuity → derivatives and local approximation → Riemann integration and FTC → infinite series and power series.
R25.05–R25.08 remain separate Atlas jobs: multivariable calculus, completeness at greater depth, measure and Lebesgue integration, and the controlled exchange of limiting operations. This page does not pre-empt those later owners.
Sources and further study
Use MIT OpenCourseWare 18.100A Real Analysis for a rigorous real-analysis route. The earlier BTT R25 guides provide the limit, differentiation and integration prerequisites used throughout this article.