Who made calculus rigorous?

In the 1660s–1680s, two people independently built calculus out of thin air. Isaac Newton got there first (around 1666, during the plague years) but barely published; he called his rates of change fluxions and thought in terms of quantities flowing through time. Gottfried Wilhelm Leibniz arrived a decade later from a completely different angle, thinking about infinitely small differentials — and he published first, in 1684.

Leibniz also gave us the notation you still type today: the dy/dx for a derivative and the elongated-S ∫ for an integral both come from him, not Newton. If you've ever written ∫ f dx, you're using a 340-year-old API designed by a man who thought in infinitesimals.

The two ideas were the same idea. Fluxions and differentials are two dialects for one machine: differentiation and integration as inverse operations, tied together by what we now call the Fundamental Theorem of Calculus. And it worked — spectacularly. Planetary orbits, tides, the shape of a hanging chain, the trajectory of a cannonball: calculus predicted them all. Nobody was going to stop using a tool this powerful just because the fine print was missing.

So what was the machine actually built on? A number that couldn't quite decide whether it existed. An infinitesimal — Leibniz's dx — was supposed to be a quantity that's bigger than zero but smaller than every positive number you can name. Zero-ish, but not zero. Both at once, whenever it was convenient.

Watch it in action. To differentiate y = x², you nudge x by an infinitesimal dx and expand:

(x + dx)² − x² = 2x·dx + dx²   ⟹   slope = (2x·dx + dx²) / dx = 2x + dx → 2x

Notice the sleight of hand. In the last step you divide by dx — which is only legal if dx ≠ 0. Then you drop the leftover dx — which is only legal if dx = 0. Same symbol, treated as nonzero in one breath and zero in the next. The answer, 2x, is exactly right. The reasoning is a contradiction.

This is the original sin of calculus: a tool that gave perfect answers through a step that, examined closely, divides by something you then declare to be nothing. It was powerful. It was also, logically, a bug that happened to return the correct value every time.

In 1734 someone finally said it out loud. Not a mathematician — a philosopher-bishop, George Berkeley, in a pamphlet with the gloriously combative title The Analyst: A Discourse Addressed to an Infidel Mathematician. (The "infidel" was probably Newton's friend Edmond Halley, of comet fame, who had reportedly needled a dying man for believing Christian mysteries while cheerfully accepting mathematical ones.)

Berkeley's move was ruthless: turn the mathematicians' own standard of rigor against them. If you scientists demand airtight logic before believing in God, he argued, then look at what you'll swallow in calculus. That dx you divide by and then discard — what is it?

That phrase — the ghosts of departed quantities — is the most famous line in the history of mathematical criticism. And notice: Berkeley didn't prove calculus wrong. He proved it unjustified. The answers were right; the reasons were haunted. For the next hundred years, every serious mathematician knew there was a ghost in the machine.

The first real exorcism came from Augustin-Louis Cauchy in 1820s Paris. His textbook Cours d'analyse (1821) made one decisive strategic choice: stop talking about infinitesimals as objects, and talk about limits as a process instead.

The derivative isn't a ratio of two ghosts dy/dx. It's the limit that the ordinary, honest ratio Δy/Δx approaches as Δx shrinks toward zero — where Δx is a genuine nonzero number the whole time, never asked to be zero and nonzero at once. No division by nothing. The ghost is replaced by a verb: approaching.

This is the exact idea you met back in Chapter 2: a limit isn't "the value at the end," it's "the value you can force the output as close to as you like." Cauchy was the first to build all of calculus on that foundation and to write down serious, near-modern definitions of both limit and continuity in that language — the direct ancestors of the ε–N and ε–δ statements from Chapters 2 and 3.

He wasn't quite all the way there — Cauchy still leaned on phrases like "approaches indefinitely" and occasionally slipped back into infinitesimal talk. But the strategy was set. The limit, not the infinitesimal, was now the load-bearing concept. The ghost had been given an eviction notice.

Karl Weierstrass finished the job in the 1860s. Where Cauchy still said "approaches," Weierstrass replaced every last hand-wave with pure inequalities and quantifiers. The ε–δ definition you learned in Chapter 3 — for every ε > 0 there exists a δ > 0 such that… — is his, essentially word for word. No motion, no time, no "approaching," no ghosts. Just static logic about numbers: name a tolerance, I'll name a neighborhood. That's the definition still printed in every analysis textbook today, 160 years later.

And then, having built the airtight foundation, Weierstrass used it to blow up everyone's intuition. In 1872 he exhibited a function that is continuous everywhere but differentiable nowhere — an unbroken curve with no smooth spot at any point, a coastline that's all corners, jagged at every magnification.

This was a scandal. For two centuries "continuous" had quietly been assumed to mean "smooth enough to have a tangent almost everywhere" — you draw it without lifting your pen, so surely it has a slope most places? Weierstrass's monster killed that assumption stone dead. Continuity (Chapter 3) and differentiability (Chapter 4) are genuinely different properties, and the gap between them is enormous. The only way to even state such a beast precisely, let alone prove it exists, is with ε–δ. The rigor wasn't pedantry — it was the only flashlight that could find the monster.

There was still one hole left — the deepest one. Cauchy and Weierstrass had built everything on limits of real numbers. But nobody had ever said what a real number actually is. The whole edifice rested on ℝ, and ℝ was undefined. You can't prove "every bounded sequence has a limit" if you can't say what the limit is made of.

Richard Dedekind closed it in 1872 with the construction you already met in Chapter 1: a real number is a cut of the rationals — a clean split of ℚ into a lower heap and an upper heap. Where the split lands in a gap with no rational (like √2), the cut itself is the new number. No infinitesimals, no ghosts, no appeal to geometry — just sets of fractions. Completeness stopped being an assumption and became a provable fact about a thing you could actually construct.

And then Georg Cantor, that same decade, asked a question nobody had thought to ask: how many reals are there, compared to rationals? Both are infinite — but in 1874 Cantor proved they are not the same infinite. His diagonal argument shows that any list claiming to enumerate all the reals must miss one: there are strictly more reals than rationals. ℝ is uncountable; ℚ is only countable. Infinity comes in sizes.

Sit with that. The rationals — every fraction that will ever exist — can in principle be lined up and counted 1, 2, 3, …. The reals cannot. No matter how clever your enumeration, the diagonal trick builds a real you forgot. The number line you've stood on for six chapters is vastly bigger than the fractions dotted through it — the holes outnumber the pins, uncountably.

Two centuries on one line. Each dot is a moment the foundation got firmer (or, in Berkeley's case, a moment someone proved it wasn't firm at all). Click, tap, or tab to a dot to see who did what — and exactly which crack they filled.

No Scala or Haskell here — history doesn't compile. Instead, the quick-reference card: the whole cast, their dates, and the one thing each is remembered for.

Here's the reason this history belongs at the end of the book rather than the start: you now know the definitions well enough to appreciate who fought to get them. Every airtight idea you've been handed was somebody's hard-won repair of Berkeley's ghost. The map:

That's the whole book, and the whole story. Calculus was a machine that worked for 150 years on faith, got publicly caught running on ghosts, and was rebuilt — plank by plank, limit by limit, cut by cut — into the airtight subject you just finished. Newton and Leibniz gave us the power. Berkeley gave us the bug report. Cauchy, Weierstrass, Riemann, Dedekind, and Cantor closed the tickets.

Every time you write lim, take a derivative, or trust that a bounded increasing sequence converges, you're standing on two centuries of debugging. Now you know whose names are in the commit log.