Absolute vs conditional convergence: Visual walkthrough

Before we touch symbols, let us agree on what every word means.

Step 1 — Picture a signed series as a walk

WHAT. We lay the terms $a_{n}$ on a number line as steps: a positive term steps right, a negative term steps left, and the length of the step is $∣ a_{n} ∣$ .

WHY. A sum is abstract; a walk is concrete. Convergence of $\sum a_{n}$ becomes the visible question: does the walker settle at a spot? This is the object we must control.

PICTURE. Below, the blue path is the signed walk $\sum a_{n}$ . Notice how left-steps (orange) partially undo right-steps (blue) — that undoing is cancellation, and it is the walker's friend.

Figure — Absolute vs conditional convergence

$running total after N terms S_{N} = N signed steps a_{1} + a_{2} + \dots + a_{N}$

$S_{N}$ — where the walker stands after $N$ steps.
each $a_{k}$ — one signed step (sign = direction, magnitude $∣ a_{k} ∣$ = length).

Step 2 — The "no-cancellation" walk always goes further

WHAT. Alongside the signed walk, draw a second walk where every step goes right, of the same length $∣ a_{n} ∣$ . Its total after $N$ steps is $T_{N} = ∣ a_{1} ∣ + \dots + ∣ a_{N} ∣$ .

WHY. Removing signs forbids any undoing. So the absolute walk $T_{N}$ climbs at least as fast as the signed walk drifts — it is an upper bound on how far things can move. If even this relentless climb settles down, the wilder signed walk has no room to misbehave.

PICTURE. Blue = signed walk (can go back). Orange = absolute walk (only forward). The orange curve is always above the distance travelled by blue.

$T_{N} = \sum_{n = 1}^{N} ∣ a_{n} ∣ versus S_{N} = \sum_{n = 1}^{N} a_{n}$

$T_{N}$ — total of the sign-stripped steps (never decreases).
The hypothesis of the theorem is that $T_{N}$ settles, i.e. $\sum ∣ a_{n} ∣$ converges.

We are given the orange walk converges. We must deduce the blue one does. Steps 3–6 build the bridge.

Step 3 — Manufacture a non-negative quantity to compare

WHAT. For each term define $b_{n} = a_{n} + ∣ a_{n} ∣$ . Claim: $0 \leq b_{n} \leq 2∣ a_{n} ∣$ .

WHY. Our only convergence-detecting machine for signed series that we trust from pictures is the Comparison Test — but comparison only works on non-negative terms (you can't sandwich something that jumps both ways). So we cook up a quantity that is guaranteed $\geq 0$ , yet built out of $a_{n}$ . That is $b_{n}$ .

PICTURE. Two cases decide everything:

If $a_{n} \geq 0$ (right step): $b_{n} = a_{n} + a_{n} = 2∣ a_{n} ∣$ — the top of the sandwich.
If $a_{n} < 0$ (left step): $b_{n} = a_{n} + (- a_{n}) = 0$ — the bottom of the sandwich.

The green bars ( $b_{n}$ ) are always caught between the floor $0$ and the ceiling $2∣ a_{n} ∣$ .

$0 \leq b_{n} a_{n} + ∣ a_{n} ∣ \leq ceiling 2∣ a_{n} ∣$

$b_{n} = a_{n} + ∣ a_{n} ∣$ — the manufactured non-negative term.
Lower bound $0$ — hit exactly when the step points left.
Upper bound $2∣ a_{n} ∣$ — hit exactly when the step points right.

Step 4 — Comparison Test crushes the green series

WHAT. Because $0 \leq b_{n} \leq 2∣ a_{n} ∣$ and $\sum 2∣ a_{n} ∣ = 2 \sum ∣ a_{n} ∣$ converges (given, times a constant), the Comparison Test says $\sum b_{n}$ converges.

WHY. Comparison's picture: if a taller stack of non-negative bars ( $2∣ a_{n} ∣$ ) has a finite total area, any shorter stack of non-negative bars ( $b_{n}$ ) sitting underneath it also has finite area. Finite area ↔ the running total settles.

PICTURE. Orange bars = ceiling $2∣ a_{n} ∣$ (finite total, given). Green bars = $b_{n}$ , each no taller. The green area cannot exceed the orange area, so it too is finite.

$\sum 2∣ a_{n} ∣ converges b_{n} \leq 2∣ a_{n} ∣, b_{n} \geq 0 \sum b_{n} converges$

Left side — the known convergent tower (twice the absolute series).
Arrow condition — the two hypotheses comparison demands: bounded above and non-negative.
Right side — the newly earned fact: the green series settles.

Step 5 — Undo the trick: recover the signed sum

WHAT. Solve $b_{n} = a_{n} + ∣ a_{n} ∣$ for $a_{n}$ : subtract $∣ a_{n} ∣$ from both sides to get $a_{n} = b_{n} - ∣ a_{n} ∣$ .

WHY. We invented $b_{n}$ only to sneak past comparison. Now we peel the disguise off and get the original signed term back — but as a difference of two things we now know both converge.

PICTURE. Green tower $\sum b_{n}$ (converges, Step 4) minus orange tower $\sum ∣ a_{n} ∣$ (converges, given) equals the blue signed series.

$a_{n} = converges (Step 4) b_{n} - converges (given) ∣ a_{n} ∣$

$b_{n}$ — settled by Step 4's comparison.
$∣ a_{n} ∣$ — settled by hypothesis.
Their difference is $a_{n}$ : the thing we wanted all along.

Step 6 — Difference of convergents converges: done

WHAT. Convergent series form a vector space: if $\sum b_{n}$ and $\sum ∣ a_{n} ∣$ both converge, so does $\sum (b_{n} - ∣ a_{n} ∣) = \sum a_{n}$ . $■$

WHY. This is the algebra of limits made visual: if two running totals each home in on a fixed height, their difference homes in on the difference of those heights. No new cancellation magic needed — just subtract two settled numbers.

PICTURE. As $N \to \infty$ , the green total approaches height $G$ , the orange total approaches height $A$ ; the blue signed total therefore approaches $G - A$ , a single fixed number.

$lim_{N \to \infty} S_{N} = (lim \sum b_{n}) - (lim \sum ∣ a_{n} ∣) = G - A (finite)$

$G$ — the settling height of the green series.
$A$ — the settling height of the absolute series.
$G - A$ — the guaranteed finite limit of $\sum a_{n}$ . The theorem is proved.

Step 7 — The degenerate & edge cases (never skipped)

WHAT / WHY / PICTURE for the corners the derivation must survive:

The one-picture summary

Everything above, compressed: build $b_{n}$ between floor and ceiling → comparison tames $b_{n}$ → subtract $∣ a_{n} ∣$ → get $a_{n}$ .

Recall Feynman: retell the whole walk in plain words

I want to prove that a list of pluses and minuses adds up to a real number. The trouble: minuses make the running total bounce, and my only reliable "does-it-settle?" tester needs everything pointing the same way. So I play a trick. To every term I add its own length: a right-step doubles, a left-step cancels to zero. Now every number is $\geq 0$ — my tester is allowed. And each of these fixed-up numbers is no bigger than twice the length of the original step. I was already told the tower of lengths settles, so twice it settles, so — by comparison — my fixed-up tower settles too. Finally I peel the trick off: the original term is just (fixed-up term) minus (length). Two towers that each settle, subtracted, give a tower that settles. So the signed sum lands on one number. The reason absolute convergence is stronger: if the lengths themselves settle, I never even needed the cancellation — the sum is armored. If they don't settle (like $\sum \frac{1}{n}$ ), the sum survives only by the delicate left-right dance, and shuffling the steps can send it anywhere.

Recall

In one sentence, why do we add $∣ a_{n} ∣$ to $a_{n}$ in Step 3? ::: To manufacture a non-negative quantity $b_{n}$ so the Comparison Test (which needs non-negative terms) becomes legal. Which single hypothesis powers the whole proof? ::: That $\sum ∣ a_{n} ∣$ converges (so $\sum 2 ∣ a_{n} ∣$ converges and bounds $b_{n}$ ). Why is the converse false, in picture language? ::: The signed walk can settle purely through left-right cancellation while the length-tower runs to infinity — that is conditional convergence.

Connections

Parent: Absolute vs Conditional Convergence
Comparison Test — the engine of Step 4
Alternating Series Test — proves the converse-counterexample converges
p-series · Harmonic series — why the counterexample's length-tower diverges
Riemann Rearrangement Theorem — the payoff of the absolute/conditional distinction
Ratio Test · Power series & radius of convergence — tests that detect absolute convergence directly