Infinite Series

Walk towards a wall like this: first cover half the distance. Then half of what's left. Then half of what's left after that. Your steps have lengths

\frac12 + \frac14 + \frac18 + \frac1{16} + \cdots

— infinitely many of them. Can infinitely many positive numbers really add up to something finite? It sounds impossible: surely piling on term after term, forever, must blow past every bound. And yet you plainly do reach the wall, which is exactly one room-width away. Something here adds up to 1.

But hold on — what could it even mean to add up infinitely many numbers? You cannot perform infinitely many additions; no one has that kind of time. The escape is a beautifully honest dodge: add finitely many, watch where the running total is heading, and define the infinite sum as that destination — a limit of a sequence. Let's watch the running totals of the halving series:

s_1 = \frac12, \qquad s_2 = \frac12 + \frac14 = \frac34, \qquad s_3 = \frac34 + \frac18 = \frac78, \qquad s_4 = \frac78 + \frac1{16} = \frac{15}{16}.

A pattern is staring at us: each total falls short of 1 by exactly the size of the last piece added. After N pieces,

s_N = 1 - \frac{1}{2^N},

and the gap \tfrac{1}{2^N} halves at every step. The totals never reach 1, never exceed 1, and get as close to 1 as anyone could demand. So s_N \to 1, and we declare — as a definition, not a discovery of some mystical completed addition —

\frac12 + \frac14 + \frac18 + \cdots \;=\; 1.

Around 450 BC the Greek philosopher Zeno argued that motion is impossible: to cross a room you must first cross half of it, then half the remainder, then half of that — infinitely many stages, and "surely" no one can complete infinitely many things. The paradox rattled philosophers for over two thousand years.

The modern reply fits on one line: \sum_{n=1}^{\infty} \tfrac{1}{2^n} = 1. Infinitely many stages, finite total. If each stage covers half the remaining distance, it also takes (at steady walking speed) half the remaining time — so the infinitely many stages fit snugly inside a perfectly ordinary couple of seconds. Zeno's mistake was to conflate "infinitely many steps" with "infinite total". Series are precisely the tool that tells those two apart.

The definition: a series is one sequence in disguise

Now the general story, and it is exactly the halving story with the numbers abstracted away. Given a sequence of terms a_1, a_2, a_3, \dots, build the partial sums — the running totals after 1, 2, 3, \dots terms:

s_N = a_1 + a_2 + \dots + a_N = \sum_{n=1}^{N} a_n.

Each s_N is a perfectly ordinary finite sum — no infinity anywhere yet. The infinity enters only at the last moment, through a limit. The infinite series \sum_{n=1}^{\infty} a_n is defined to be the limit of the partial sums, when that limit exists:

\sum_{n=1}^{\infty} a_n \;:=\; \lim_{N \to \infty} s_N.

If (s_N) converges to a number S, the series converges and we write \sum a_n = S. If (s_N) diverges, so does the series — the infinite sum simply has no value. Read that again, because it is the whole concept: the series converges if and only if the sequence of partial sums converges. Every question about a series — does it converge? to what? how fast? — is secretly a question about one sequence, (s_N). Everything you proved about limits of sequences now works for series, free of charge.

The dictionary runs both ways, too: the sequence recovers the terms. Since s_N and s_{N-1} differ by exactly the newest term,

a_N = s_N - s_{N-1}.

This little identity has a famous consequence. If \sum a_n converges, then s_N \to S and s_{N-1} \to S as well, so a_N = s_N - s_{N-1} \to S - S = 0. The terms of a convergent series must shrink to zero. Flip that around and you get a cheap, ruthless test for divergence.

For a sequence of terms (a_n), define the partial sums s_N = \sum_{n=1}^{N} a_n. Then the series \sum_{n=1}^{\infty} a_n is defined by

\sum_{n=1}^{\infty} a_n = \lim_{N \to \infty} s_N,

Three traps snare almost every newcomer to series:

A telescoping series, line by line

Time to see the definition earn its keep on a series that isn't just halving. The cleanest convergent series to evaluate completely by hand is the telescoping one, where consecutive terms cancel and the partial sum collapses to almost nothing — like the sliding sections of an old sailor's telescope. We compute

\sum_{n=1}^{\infty} \frac{1}{n(n+1)} = \frac{1}{1\cdot 2} + \frac{1}{2\cdot 3} + \frac{1}{3\cdot 4} + \cdots = \frac12 + \frac16 + \frac1{12} + \cdots

Step 1 — Split each term by partial fractions. Look for constants with \tfrac{1}{n(n+1)} = \tfrac{A}{n} + \tfrac{B}{n+1}. Clearing denominators gives 1 = A(n+1) + Bn; setting n = 0 yields A = 1, and n = -1 yields B = -1. So

\frac{1}{n(n+1)} = \frac{1}{n} - \frac{1}{n+1}.

Step 2 — Write out the partial sum with the split terms. This is where the definition points us: don't stare at the infinite sum, compute s_N.

s_N = \sum_{n=1}^{N}\left(\frac{1}{n} - \frac{1}{n+1}\right) = \left(\frac11 - \frac12\right) + \left(\frac12 - \frac13\right) + \cdots + \left(\frac1N - \frac1{N+1}\right).

Step 3 — Watch the middle collapse (the telescope). Every -\tfrac1k is cancelled by the +\tfrac1k in the next bracket. All that survives is the very first term and the very last:

s_N = \frac11 - \frac{1}{N+1} = 1 - \frac{1}{N+1}.

Sanity-check it: s_1 = \tfrac12, s_2 = \tfrac12 + \tfrac16 = \tfrac23, s_3 = \tfrac23 + \tfrac1{12} = \tfrac34 — and indeed 1 - \tfrac12, 1 - \tfrac13, 1 - \tfrac14. The formula holds.

Step 4 — Take the limit of the partial sums. As N \to \infty, the leftover tail \tfrac{1}{N+1} \to 0 (it is the harmonic sequence shifted by one), so

\sum_{n=1}^{\infty} \frac{1}{n(n+1)} = \lim_{N\to\infty} s_N = \lim_{N\to\infty}\left(1 - \frac{1}{N+1}\right) = 1. \qquad \blacksquare

Infinitely many positive terms, and yet they total exactly 1 — because each new term is small enough that the running total is reined in. Notice how the whole argument was just the definition, executed: find a closed formula for s_N, take a limit.

The same collapse works from any starting point. Begin the sum at n = k instead of n = 1 and the surviving first term is \tfrac1k, so the tail of the series telescopes to

\sum_{n=k}^{\infty} \frac{1}{n(n+1)} = \frac{1}{k}.

Tails like this — "everything from the k-th term onward" — become the main characters later, when you estimate how fast a series converges.

The shocker: the harmonic series diverges

Now for the most famous cautionary tale in the subject. The harmonic series is the sum of the reciprocals,

\sum_{n=1}^{\infty} \frac{1}{n} = 1 + \frac12 + \frac13 + \frac14 + \frac15 + \cdots

Its terms \tfrac1n \to 0, just like the telescoping series' terms. Its early partial sums crawl: s_4 \approx 2.08, s_{10} \approx 2.93. Everything about it looks convergent. It is not. The partial sums climb past every ceiling — 10, 100, a trillion — given enough terms. Here is the block argument, and it needs nothing but grouping.

Bundle the terms in blocks whose sizes double:

1 + \underbrace{\frac12}_{} + \underbrace{\frac13 + \frac14}_{2\text{ terms}} + \underbrace{\frac15 + \frac16 + \frac17 + \frac18}_{4\text{ terms}} + \underbrace{\frac19 + \cdots + \frac1{16}}_{8\text{ terms}} + \cdots

Replace every term in a block by the smallest one in it (the last). Then \tfrac13 + \tfrac14 > \tfrac14 + \tfrac14 = \tfrac12, and \tfrac15 + \dots + \tfrac18 > 4\cdot\tfrac18 = \tfrac12, and \tfrac19 + \dots + \tfrac1{16} > 8\cdot\tfrac1{16} = \tfrac12 — every block exceeds \tfrac12:

\sum_{n=1}^{\infty} \frac1n \;>\; 1 + \frac12 + \frac12 + \frac12 + \cdots = \infty.

Adding \tfrac12 infinitely often is unbounded, so the partial sums climb past every ceiling — slowly, but inexorably. Terms shrinking to zero is necessary for convergence but never sufficient: whether a series converges depends not just on the terms dying out, but on how fast. The terms \tfrac{1}{n(n+1)} \approx \tfrac{1}{n^2} die fast enough; the terms \tfrac1n do not, by the narrowest of margins.

That doubling-blocks argument is not modern. It was found by Nicole Oresme — a French philosopher, economist and eventually bishop — around 1350, three centuries before calculus existed. His proof was then lost to history and the result had to be rediscovered in the 1600s by Pietro Mengoli (the same Mengoli who first summed our telescoping series) and the Bernoulli brothers.

Nearly seven centuries later, Oresme's argument is still the one every textbook teaches — it has never been beaten for sheer economy. Not bad for the Middle Ages: no limits, no notation, no \Sigma — just the observation that you can always scoop up enough tiny terms to make another half.

Settling versus running away

A series lives or dies by its partial sums s_N. Plot them as dots and the verdict is visible: a convergent series' dots level off toward a horizontal line (its sum), while a divergent series' dots keep climbing with no ceiling. Switch between the telescoping series \sum \tfrac{1}{n(n+1)} (settling to 1, drawn dashed) and the harmonic series \sum \tfrac1n (running away).

A word of warning about reading such pictures: the harmonic dots climb ever more slowly, and over any window they look like they're levelling off. They aren't. The harmonic partial sums first pass 10 at N = 12{,}367, and to pass 100 you would need roughly 10^{43} terms — more terms than there are grains of sand on every beach of a billion Earths. No computer plot can tell "converges" from "diverges at a glacial pace". Only the mathematics can: convergence is about the destination of (s_N), not the speed of travel, and that is why the limit definition — not the picture — gets the final word.

See it explained