Series and Convergence Intro

An infinite sum converges when its running totals settle toward one number, which happens when new terms shrink fast enough to stop mattering.

Partial sums 1, 1.5, 1.75 climb toward the ceiling 2 in halving steps, showing convergence as gaps that shrink to nothing

Why Does This Exist?

Infinite sums look like a joke: add forever, get... a number? Yet Taylor approximations, expected values, and iterative algorithms all are infinite processes that must land somewhere. Convergence is the landing test: do the running totals approach a finite limit, or do they escape to infinity or oscillate forever? Shipping an "infinite sum" that diverges is how models output inf and proofs collapse.

The good news for an intro: one family, geometric series, covers the intuition completely, and one rule of thumb (terms must shrink fast, not just shrink) prevents the famous mistake. Listing versus accumulating is in sequences and series; the working families are in arithmetic and geometric sequences.

The one idea you need first

A series converges to $L$ when its partial sums $S_N$ get and stay arbitrarily close to $L$ as $N$ grows: for any tiny tolerance, all sufficiently late totals sit inside it. Otherwise it diverges (growing unbounded, oscillating, or wandering). Convergence is a property of the totals, never of single terms.

Think of It Like This

Walking half the remaining distance

Stand 2 meters from a wall and repeat: walk half the remaining gap. First stride 1 meter, then half a meter, then a quarter. You close in relentlessly yet each stride is smaller than the last, and the wall is never passed. The total walked converges to 2.

Now stride a full meter every time instead: 1, 2, 3, past the wall into the next room. Same step count, no settling. Convergence is not about taking many steps but about steps shrinking fast enough that their total stays bounded. The analogy stops at oscillation: no walking pattern captures totals that swing forever without settling.

How It Actually Works

The geometric series with ratio $r$ is the reference case:

\sum_{n=0}^{\infty} r^n = 1 + r + r^2 + \cdots = \frac{1}{1 - r}, \qquad |r| < 1

In plain words: each term multiplies the last by $r$ ; when $|r|$ (absolute value) is below 1, terms decay exponentially and the totals settle at $1/(1-r)$ . When $|r| \geq 1$ the series diverges (terms refuse to vanish, or grow). The $r = 1/2$ case sums to $1/(1 - 1/2) = 2$ .

The essential caution: terms shrinking to zero is necessary but NOT sufficient. The harmonic series $1 + 1/2 + 1/3 + \cdots$ has vanishing terms yet diverges (its totals pass 3, 4, 5, ... logarithmically). Speed of shrinking decides, not shrinking alone.

Worked example

Sum $1 + 1/2 + 1/4 + 1/8 + \cdots$ ( $r = 1/2$ ):

$S_1 = 1$ ; $S_2 = 1.5$ ; $S_3 = 1.75$ ; $S_4 = 1.875$ ; gap to 2 halves each step: $1, 0.5, 0.25, 0.125$ .
Formula: $1/(1 - 1/2) = 2$ . At $N = 6$ , $S_6 = 1.96875$ , within $0.04$ of the limit.
Contrast: harmonic partial sums at $N = 4$ reach $2.083$ and keep creeping past every bound eventually. Vanishing terms, diverging totals.

Same "terms go to zero" appearance, opposite fates. The ratio test ( $|r| < 1$ converges) separates them instantly here.

Watch Out For

Assuming small terms mean a finite total

Truncating the harmonic series at a million terms gives about 14.4, which looks settled until the next billion terms add 7 more. "Terms are tiny so the tail is negligible" is the costliest convergence mistake: tails of slowly-decaying series are enormous.

Bound the tail with a test (integral, comparison, ratio) before truncating. If no test applies, say the tail is unknown rather than small.

Summing a divergent series numerically and trusting it

Partial sums of $1 + 2 + 4 + 8 + \cdots$ overflow to inf while alternating series like $1 - 1 + 1 - \cdots$ bounce 1, 0, 1, 0 forever. Code that loops "until stable" either crashes or returns whichever bounce it happened to stop on.

Check convergence analytically before summing numerically. If totals oscillate or explode within the first dozen terms, stop: more terms will not rescue them.

The Quick Version

Convergence means partial sums approach one finite limit; otherwise the series diverges.
Geometric series with $|r| < 1$ sum to $1/(1-r)$ : halving terms total exactly 2.
$S_3 = 1.75$ with gaps halving each step toward the ceiling.
Terms tending to zero is necessary but not sufficient: the harmonic series diverges.
Bound tails with a real test before truncating any infinite sum.
Never sum numerically before establishing convergence on paper.