Calculus Didn't Start With a Textbook, It Started With Someone Trying to Catch a Tortoise

Nobody wakes up one day and decides to invent calculus for fun. At its core, the whole subject is just the mathematics of continuous change, how something moves, grows, or accumulates when it's doing so constantly, rather than in neat, countable steps. Every piece of it, the derivative, the integral, the differential equation, the infinite series, was built because someone got stuck on a problem that no existing mathematics could touch: how fast is something moving at this exact instant, what's the exact area under a curve that refuses to be a triangle or a circle, what does something look like if all you know is its rate of change, how do you calculate the value of a function you can't actually compute by hand. If calculus has ever felt like a pile of symbols someone handed you without an explanation, that's genuinely not your fault, it took the smartest people in the world over two thousand years to agree on what those symbols should even mean.

This article builds it back up the way it was actually built, one stuck problem at a time, starting nearly 2,500 years before anyone wrote a derivative, and following it all the way through the centuries of arguing, refining, and occasionally public humiliation it took to reach the calculus you're about to learn.

The Problem That Broke Ancient Mathematics: Zeno's Paradox

Around 450 BC, a Greek philosopher named Zeno of Elea proposed a race that should be embarrassingly easy to call. Achilles, the fastest runner in Greek myth, races a tortoise. To make it fair, the tortoise gets a head start. Zeno's argument: Achilles can never actually catch the tortoise.

Here's why, in Zeno's telling. Before Achilles can catch the tortoise, he first has to reach the point where the tortoise started. But by the time he gets there, the tortoise has crawled a bit further ahead. So Achilles now has to reach that new point too. But by the time he does, the tortoise has moved a little further still. This repeats forever, an infinite number of catch-up points, each one requiring Achilles to close a gap before the next tiny gap opens up. Infinite steps, Zeno reasoned, must take infinite time. Achilles never wins.

You already know this is wrong. Fast runners overtake slow ones constantly, you've probably seen it happen in an actual race. The unsettling part isn't the conclusion, it's that for a very long time, nobody could immediately say why the logic was flawed. If it feels a little unsettling to you too, that's a good sign, not a bad one, it means you're taking the argument seriously enough to notice it should be wrong.

The problem sat right at the boundary of something Greek mathematics had no good tools for: infinite processes. You could add up two numbers, or ten, or a thousand. Could you meaningfully add up infinitely many numbers and get a normal, finite answer? Nobody knew. That gap in the mathematics would stay open for nearly two thousand years, and closing it, eventually, is most of what this article is about.

Archimedes Nearly Invents Calculus, 2,000 Years Early

The first real crack in that gap came from Archimedes, working around 225 BC in Syracuse. Archimedes wanted to find the exact area of a parabolic segment, the curved region cut off by a straight line slicing through a parabola. There was no formula for this. Circles and triangles had formulas. Curves didn't, and if you've only ever measured area with a formula, it's worth sitting with how strange that gap must have felt.

His method, building on an approach called exhaustion that mathematicians going back to Eudoxus had been refining since around 370 BC, was to fill the curved region with triangles. Draw one big triangle inside the segment. It doesn't fill the whole area, there's a smaller curved sliver left on each side. Draw two more triangles in those slivers. Still more slivers left, smaller this time. Draw four more triangles. Keep going, forever.

Archimedes showed that each new round of triangles adds up to exactly one quarter of the area added in the previous round. If the first, biggest triangle has area T, the total area of the parabolic segment is:

T + T/4 + T/16 + T/64 + ... (continuing forever)

This is an infinite sum, exactly the kind of thing Zeno's paradox suggested might be meaningless. Watch what happens when you actually add it up:

Add infinitely many shrinking triangles and you land, exactly, on four-thirds of the first triangle's area. Not approximately. Exactly. Here's why it's exactly 4T/3, and not just a number the table happens to be creeping toward. Factor out the T and call the rest S, so S = 1 + 1/4 + 1/16 + 1/64 + ..., forever. This is what's called a geometric series, each term is the one before it multiplied by a fixed ratio, in this case 1/4. There's a trick for these that turns an infinite, unfinishable sum into a one-line piece of ordinary algebra.

Multiply the whole series by that same ratio, 1/4:

S/4 = 1/4 + 1/16 + 1/64 + 1/256 + ...

Now look closely at S and S/4 side by side. Every term in S/4 also appears in S, just shifted over by one position. Subtract S/4 from S, and every one of those shared terms cancels out perfectly, all the way to infinity, leaving only the very first term of S behind:

S − S/4 = 1 (3/4)S = 1 S = 4/3

No approximation anywhere in that argument, no trailing off with "and so on." It's a finite, three-line proof that an infinite sum equals exactly 4/3, which means the parabolic segment's area is exactly 4T/3, full stop.

This is also worth remembering as a reusable tool, not just a one-off trick for this problem: any geometric series with first term a and ratio r, as long as r is between −1 and 1 so the terms genuinely shrink, sums to exactly a/(1−r). Plug in a = 1, r = 1/4, and you get 1/(1−1/4) = 1/(3/4) = 4/3, the same answer, arrived at through the general formula instead of the one-off subtraction trick.

That formula shows up again later in this article, when Madhava's infinite series for sine and cosine get built the same way, and it's the mathematical reason Zeno's paradox, despite sounding airtight, was never actually describing a real problem, an infinite number of shrinking steps can add up to a perfectly finite, exact distance, which is exactly what lets Achilles catch the tortoise on schedule.

What Archimedes didn't have was a general method. Every new curve needed its own custom-built geometric trick. There was no algebra of change, no reusable machine you could point at any curve and crank out an answer. That machine wouldn't get built in Europe for another eighteen centuries. Somewhere else in the world, though, it very nearly did, and far earlier than most calculus textbooks ever mention.

Meanwhile, Centuries Before Europe Caught Up: Madhava of Sangamagrama

Around 1340, in the town of Sangamagrama in what's now Kerala, in southern India, a mathematician-astronomer named Madhava was born into a problem that had nothing to do with Zeno or Archimedes at all, he needed increasingly precise tables of the sine function for astronomical calculations, and existing methods only got him approximate values.

Madhava's solution, developed and refined by the school of mathematicians and astronomers he founded, was to represent the sine and cosine functions, and later the arctangent function, as infinite sums, expressions built from adding together an endless string of shrinking terms, similar in spirit to Archimedes' shrinking triangles but aimed at a completely different kind of problem. As a special case of his arctangent series, Madhava derived an infinite series for calculating π itself, and used it to compute the value correct to somewhere between eleven and thirteen decimal places, an extraordinary feat for the 14th century, done entirely by hand.

Here's the series itself, arguably the simplest and most famous piece of his work, and worth actually running to see how it behaves:

π/4 = 1 − 1/3 + 1/5 − 1/7 + 1/9 − 1/11 + ...

The true value of π/4 is about 0.7854. The partial sums are visibly oscillating around that target, above it, then below it, then above again, homing in, but doing so painfully slowly, still off by about 0.03 after eight terms. This is exactly why Madhava didn't stop at the raw series: he went on to work out correction terms that estimate the remaining error after cutting the sum short, letting a handful of terms plus a correction do the job that thousands of raw terms would otherwise take. That refinement, on top of an already correct infinite series, is what let him reach eleven to thirteen decimal places of π by hand in the 14th century.

This series for π resurfaced in Europe roughly three centuries later, first as a partial rediscovery by the Scottish mathematician James Gregory in 1671, then independently by Leibniz around 1675, who had no knowledge Madhava's work existed. It's still sometimes called the Gregory-Leibniz series in Western textbooks, even though it properly belongs to a mathematician working nearly three hundred years earlier and half a world away.

It's worth being precise about what this does and doesn't mean, so nobody walks away with the wrong idea: Madhava and his successors in the Kerala school never combined their series work with a general theory connecting rates of change and areas the way Newton and Leibniz eventually would, they didn't build the unifying framework that turns calculus into a general problem-solving tool rather than a brilliant set of individual results. But the specific mathematical object, an infinite series standing in for a function too complicated to compute directly, existed and worked, correctly, centuries before anyone in Europe touched the idea. Keep Madhava in mind. He comes back later in this article, once there's enough machinery built up to see exactly why his idea matters so much.

The 17th Century Needed New Math, and Got It From Some Unlikely Places

Mathematics didn't sit still for eighteen centuries, but the specific problem of continuous, changing quantities mostly did in Europe, until a cluster of very practical, very human problems forced the issue in the 1600s.

One came from a wine cellar, which is a fairly relatable place for a math breakthrough to start. In 1613, the astronomer Johannes Kepler remarried, and while buying wine for the celebration, he noticed the merchant measuring the volume of a barrel by poking a single rod in diagonally through the bunghole and reading off one number, regardless of the barrel's actual shape.

Annoyed that this crude method seemed to work anyway, Kepler spent the next two years working out the actual mathematics of curved volumes, publishing Nova Stereometria Doliorum Vinariorum (New Solid Geometry of Wine Barrels) in 1615, treating a barrel as an infinite stack of thin circular slices and summing them, the same "chop it into thin pieces and add them up" instinct Archimedes used for triangles, just aimed at a solid shape instead of a flat area. It's a direct ancestor of what we now call integration, motivated by a mildly annoyed astronomer at a party.

Kepler had bigger fish to fry too. His laws of planetary motion, published between 1609 and 1619, showed that planets travel in ellipses, not circles, and sweep out area at a constant rate even though their speed constantly changes, moving faster near the sun and slower far from it. Describing exactly how a planet's speed changes moment to moment, along a curved path, was a problem existing mathematics simply couldn't express.

Around the same time, Pierre de Fermat, working through correspondence in the 1630s, developed something he called the method of adequality to find the tangent line to a curve at a point, and to locate a curve's maximum or minimum points. His trick: take a point on the curve, nudge it over by a tiny amount, write out the algebra for that nudge, then let the nudge shrink toward nothing and see what's left. Watch it actually work on a simple, concrete problem: split a length of 20 into two pieces, x and (20 − x), and find the value of x that makes their product as large as possible.

Fermat's move was to compare the product at x against the product at a nudged point, x + e, and treat the two as "adequal," nearly equal, right at the maximum:

x(20 − x) ≈ (x + e)(20 − x − e)

Expand the right side: (x + e)(20 − x − e) = 20x − x² + 20e − 2xe − e². Set that adequal to the left side, 20x − x²:

20x − x² ≈ 20x − x² + 20e − 2xe − e² 0 ≈ 20e − 2xe − e²

Divide through by e, then let e shrink to nothing:

0 ≈ 20 − 2x − e → 0 = 20 − 2x → x = 10

The maximum product occurs at x = 10, giving 10 × 10 = 100. That's the correct answer, and the algebra that produced it, expand, cancel the shared terms, divide by the nudge, then let the nudge vanish, is essentially identical to the derivative-based optimization built later in this article, arrived at four decades before Newton or Leibniz existed.

By the 1660s, mathematics had a pile of powerful but disconnected tricks, Archimedes' exhaustion, Madhava's series, Kepler's slicing, Fermat's vanishing nudges, and no single framework tying them together. That framework arrived from two directions at once, and the two men who built it never fully agreed on who got there first.

Newton's Plague Year, and Leibniz's Universal Language

In 1665, the Great Plague forced Cambridge University to close. Isaac Newton, then twenty-two, not much older than you probably are right now, retreated to his family's farm in Woolsthorpe, Lincolnshire, and over roughly eighteen months of enforced isolation, worked out the foundations of his law of universal gravitation, major advances in optics, his generalized binomial theorem (a way of expanding expressions like (1+x) raised to any power, including fractional and negative powers, as an infinite series, a result closely related to Madhava's series work centuries earlier and continents away, arrived at through completely independent reasoning), and a general method for handling instantaneous rates of change and areas under curves, all more or less at once.

He called his method the theory of fluxions, imagining every changing quantity as something "flowing" through time, with its fluxion being the rate of that flow, what we'd now call a derivative.

Newton didn't publish any of this for decades. His fluxion notation only appeared in print as an appendix to his 1704 Opticks, with the fuller treatment not published until 1736, nine years after his death.

Roughly a decade later, working entirely independently in Germany and unaware of Newton's unpublished notes, Gottfried Wilhelm Leibniz arrived at the same core ideas through a different route.

Where Newton was driven by physics and motion, Leibniz was chasing something more abstract, a universal symbolic language capable of representing precise reasoning about anything. Calculus, for Leibniz, was a piece of that larger project, and it shows in how carefully he built its notation. By 1675 he'd settled on dy/dx for a rate of change and ∫ for a sum (a stretched-out letter S, for the Latin summa), and published this notation between 1684 and 1686. It's the exact notation you're about to learn. Newton's dot-based fluxion notation is not.

Slope, Secants, and Where dy/dx Actually Comes From

You've almost certainly already met slope, back in coordinate geometry: rise over run, how steep a line is.

Quick check, one you could do in your head: a line passing through (1, 2) and (4, 8). Slope = (8 − 2) / (4 − 1) = 6 / 3 = 2. Pick any other two points on that same line and you'll get 2 again, guaranteed.

Here's where it gets interesting. A curve doesn't have one slope. It has a different steepness at every single point along it, steeper here, flatter there, maybe even sloping backward somewhere else. So the phrase "the slope of a curve" doesn't actually mean anything yet, on its own, until you also say where.

The first honest attempt at an answer is to do exactly what you already know how to do: pick two points on the curve and draw a straight line connecting them. This connecting line is called a secant, and its slope, using the exact same rise-over-run formula, tells you the average steepness of the curve between those two points. It's a real number, it's just not quite the answer to "how steep is the curve right here," because it's smearing the answer across a whole stretch of curve instead of pinning it to one spot.

Watch what happens on the curve y = x², fixing one point at (1, 1) and sliding the second point closer and closer to it:

As the second point slides in, the secant line isn't just changing slope randomly, it's swinging around and settling into one exact direction, the direction the curve itself is heading at (1, 1). Slide the second point close enough, and the secant line stops being an approximation and becomes the tangent line, the one line that touches the curve at (1, 1) and points in exactly the same direction the curve is pointing at that instant, without cutting across it. That tangent line's slope, 2 in this case, is the derivative at that point. If you take away one picture from this entire article, make it this one: a secant line collapsing down onto a tangent line. Everything called "instantaneous rate of change" from here on is secretly this.

This is also exactly where the notation comes from, and it's a genuinely nice piece of design, worth knowing rather than just memorizing. The Greek capital letter Δ (delta) means "a change in," so Δx means "a change in x" and Δy means "a change in y." Slope, written properly, is Δy/Δx, the same rise-over-run formula, just relabeled. That's what you're computing every time you build a secant line above.

Leibniz needed a way to write down what's left once that gap has shrunk all the way down to nothing, the tangent line's slope, without it looking like nonsense. His solution was almost poetic: swap the capital Greek Δ for a small, plain Roman d. Δx is a visible, ordinary change, something you could measure with a ruler. dx is what's left of that change once it's been shrunk infinitely small, and dy/dx is the ratio of two such vanishingly small changes, still behaving, remarkably, exactly like a fraction.

Here's the twist worth keeping, because it connects directly to a fight later in this article. Leibniz genuinely believed dx and dy were real, if unimaginably tiny, quantities, actual infinitesimal numbers you could divide one by the other. That belief is precisely what Bishop Berkeley tore apart in 1734, and precisely what Cauchy and Weierstrass eventually replaced, roughly 150 years later, with the modern, fully rigorous definition: dy/dx isn't a fraction of two tiny numbers at all, it's shorthand for the limit of Δy/Δx as Δx shrinks toward zero, exactly the secant-to-tangent process pictured above, made mathematically airtight. Leibniz's beautiful notation survived completely intact. What it technically means underneath the symbol changed entirely, a century and a half after he first wrote it down.

Limits: Making "Approaching" Precise

Before differentiation and integration can be built properly, they need a shared foundation underneath both of them, one specific idea that both Newton and Leibniz used constantly but never fully pinned down: what does it actually mean for something to "approach" a value without ever quite reaching it?

Try a small puzzle, nothing to do with derivatives yet. Consider the expression (x² − 1) / (x − 1). Plug in x = 1 directly and you get 0/0, which is meaningless, undefined, a wall. It's tempting to just stop there and call the question unanswerable. But ask instead what happens as x gets closer and closer to 1, without landing on it exactly.

From both sides, above and below, the expression is homing in cleanly on 2. Algebraically, this makes sense once you notice x² − 1 factors into (x−1)(x+1), so the expression simplifies to (x+1) everywhere except exactly at x = 1, where it's genuinely undefined, but the limit, the value it approaches, is still perfectly well defined and equals 2.

This idea, a value that a function approaches, whether or not it's actually reached, is the load-bearing concept underneath every piece of calculus that follows. Fermat's vanishing nudge, Newton's fluxion, Leibniz's infinitesimal difference, all of them are really this same idea, wearing different historical costumes. Formalizing it properly, in a way that survives serious philosophical scrutiny, turns out to be surprisingly hard, hard enough that it took mathematicians another 150 years after Newton and Leibniz to get it fully right, a story this article circles back to later. For now, the working intuition is genuinely enough to build on: a limit is where something is headed, described with total precision, even when it never technically arrives.

Building Differentiation From Scratch: The Mathematics of "Right Now"

Here's the actual problem differentiation solves, built up the way Newton and Fermat would have approached it, not the way a formula sheet presents it.

You already know how to find an average speed: distance traveled divided by time taken. Drop a ball and let it fall for 2 seconds, then measure how far it fell, divide, done. But that only tells you the average speed across those 2 seconds. What was its speed at the exact instant t = 2 seconds, one single moment with no duration at all? Average speed needs a time interval to measure across. An instant has no interval. This looks like a contradiction, and for a long time, it effectively was one, so if it trips you up for a second, you're in good company.

Here's the trick that breaks it open, using exactly the limit idea from the previous section. Suppose a falling object's position is given by s(t) = 5t² (a simplified version of the real physics, in convenient units), measuring distance fallen at time t. To find the instantaneous speed at t = 2, start by computing the average speed over a series of ever-shrinking intervals starting at t = 2:

As the interval shrinks toward zero width, the average speed isn't wandering randomly, it's homing in on a single number: 20, exactly the same "approaching a value" behavior you just watched in the limit example above. That number is what "instantaneous speed at t = 2" actually means. Not a contradiction. A limit.

That's the entire concept behind the derivative, before any formula gets involved. Formalized:

Apply this definition directly to f(x) = x², from first principles, no shortcuts, just to see the machine actually run:

f'(x) = lim (h→0) of [(x+h)² − x²] / h = lim (h→0) of [x² + 2xh + h² − x²] / h = lim (h→0) of [2xh + h²] / h = lim (h→0) of (2x + h) = 2x

So the derivative of x² is 2x. Check it against the falling-ball example above: s(t) = 5t² would have s'(t) = 10t, and at t = 2, that gives 10 × 2 = 20, exactly matching the number the shrinking intervals converged to. The formula isn't a coincidence handed down from nowhere. It's the shrinking-interval trick, done once in general instead of once per problem.

Worth watching that pattern confirm itself once more before trusting it fully. Apply the identical first-principles definition to f(x) = x³:

f'(x) = lim (h→0) of [(x+h)³ − x³] / h

Expanding (x+h)³ takes a bit more algebra, but it works out to x³ + 3x²h + 3xh² + h³, so:

= lim (h→0) of [3x²h + 3xh² + h³] / h = lim (h→0) of (3x² + 3xh + h²) = 3x²

Line the two results up side by side: x² produced 2x, and x³ produced 3x². The power drops by one each time, and the old power reappears out front as a multiplier, exactly the pattern stated below as the power rule. It's not a rule to take on faith, it's a pattern the same shrinking-interval algebra keeps producing, every single time, for any power at all:

With that table in hand, here's a genuinely useful payoff, worth keeping for the rest of this article and well beyond it. Since the derivative measures a tangent line's slope, and a tangent line's slope tells you whether a curve is climbing, falling, or momentarily flat, setting the derivative equal to zero finds exactly the points where a curve stops climbing and starts falling, or vice versa, its peaks and valleys.

Take y = −x² + 4x, a curve that rises and then falls. By the power rule just confirmed above, its derivative is y' = −2x + 4. Set that to zero: −2x + 4 = 0, so x = 2, and plugging back in, y = −(2)² + 4(2) = 4. The tangent line is perfectly horizontal at exactly (2, 4), the single highest point on the entire curve.

This is, in miniature, how calculus finds the best possible outcome of almost anything, the highest profit, the lowest cost, the fastest route, by asking a curve to point out its own flattest spot, exactly the same maximizing move Fermat's adequality made on the length-20 problem earlier, just with the shrinking-nudge algebra now packaged into one reusable rule.

Building Integration From Scratch: The Mathematics of "Total Accumulation"

Now for the mirror-image problem, Archimedes' problem, done generally instead of one clever geometric trick at a time. How do you find the exact area under a curve, like y = x², between x = 0 and x = 2, when there's no triangle or circle formula that applies?

The same shrink-it-down instinct works again, just aimed at rectangles instead of intervals. Chop the region under the curve into thin vertical rectangles, add up their areas, and watch what happens as the rectangles get thinner and more numerous.

The approximation keeps tightening toward a single number, 2.667, or exactly 8/3.

Here's why it's exactly 8/3, the same way the Archimedes triangles landed exactly on 4/3 rather than just approaching it. Split the interval from 0 to 2 into n equal rectangles, each of width 2/n, and use the right edge of each rectangle to set its height:

No approximation left standing anywhere in that chain, just like the Archimedes proof earlier. The rectangles don't just numerically creep toward 8/3 the way the table above suggests, they're provably, exactly equal to it in the limit. That limiting value is the definite integral:

Here's where calculus stops being two separate clever tricks and becomes one unified theory, and it's worth slowing down for, because this is genuinely the best part. There's a second, completely different-looking way to get 8/3: find a function whose derivative is x², then just evaluate it at the endpoints. The derivative of x³/3 is x² (check it with the power rule: 3 × x²/3 = x²). So:

x³/3 evaluated at x=2, minus x³/3 evaluated at x=0 = 8/3 − 0 = 8/3

Same answer, arrived at through what looks like an entirely unrelated method, running the derivative machinery backward instead of building up rectangles.

This connection is the Fundamental Theorem of Calculus, and it's the actual crowning insight Newton and Leibniz each independently discovered: the tangent-line problem (differentiation) and the area problem (integration) aren't two separate branches of mathematics. They're inverse operations of each other, the same way multiplication and division are. Two problems that had been chased separately for two thousand years, from Zeno's runners to Archimedes' triangles to Fermat's vanishing nudges, turned out to be two views of a single idea, which is a genuinely satisfying thing to have proven to yourself by this point.

Differential Equations: Working Backward From the Rate to the Thing Itself, and a Chain That Humbled Its Own Challenger

There's a third kind of problem calculus was built to solve, and it runs in yet another direction. Sometimes you don't know a quantity directly, and you don't want its rate of change either, you know the rate of change and want to reconstruct the quantity itself.

Take a population of bacteria that grows at a rate proportional to how large it currently is, more bacteria means more reproducing means faster growth. In symbols:

This is a differential equation, an equation involving a rate of change rather than the quantity itself, and solving it means finding the actual function y(t) that satisfies it. The technique, separation of variables, rearranges the equation to put all the y-terms on one side and all the t-terms on the other, then integrates both sides:

dy/y = k dt, then integrating both sides gives ln(y) = kt + C, which rearranges to y = Ce^(kt)

Check it yourself: differentiate y = Ce^(kt) and you get dy/dt = kCe^(kt) = ky, exactly satisfying the original equation. This single equation, exponential growth or decay depending on the sign of k, describes bacterial growth, radioactive decay, compound interest, and how a hot cup of coffee cools toward room temperature, all with the same underlying mathematics, just different constants plugged in.

This branch of calculus grew fast, and its origin story has a genuinely humbling twist. In May 1690, Jacob Bernoulli published a public challenge in Acta Eruditorum: find the exact curve traced by a flexible chain hanging freely between two fixed points. Galileo had guessed, wrongly, that it was a parabola, a mistake a sixteen-year-old Christiaan Huygens had already disproven decades earlier without offering the correct answer.

Jacob's own brother Johann, along with Huygens and Leibniz, each independently worked out the answer, a curve now called the catenary, publishing their solutions in June 1691. Jacob Bernoulli, the mathematician who had posed the challenge in the first place, failed to solve it himself in time, and had to watch his younger brother succeed where he hadn't. It's a nice reminder that even the people writing the history books were sometimes the ones stuck on the problem.

Buried inside Jacob's own notes on the problem, though, sits one more small, permanent legacy: it's the first place the word "integral" appears in print with the meaning it still carries today, coined by the man the challenge ultimately defeated.

The bacteria equation above is what's called a first-order differential equation, it only involves a rate of change, not a rate of a rate. Newton's own equation, the one that actually forced him to invent all of this, is second-order. Acceleration is the rate of change of velocity, which is itself the rate of change of position, so force = mass × acceleration involves the rate of change of the rate of change of position, written d²x/dt².

Solving for how planets actually move under gravity meant solving a differential equation, which meant Newton needed a mathematics of rates of change before he could finish the physics he actually cared about.

Second-order equations generally need more machinery than the simple separation of variables shown above, an entire further branch of the subject in its own right, but the underlying idea never changes: state a relationship between a quantity and its rates of change, then work backward to find the quantity itself. A swinging pendulum or a bouncing spring, what's called simple harmonic motion, is one of the cleanest examples, and it's governed by an equation of exactly this second-order shape. Calculus wasn't a side project for Newton. It was the tool he had to build first, and first-order equations were only ever the entry point.

Sequences and Infinite Series: Madhava's Head Start, and Taylor's Idea That Took Sixty Years to Matter

Earlier in this article, Madhava of Sangamagrama built infinite series for sine, cosine, and arctangent, three centuries before Newton and Leibniz, without the general framework of derivatives and integrals to build them from. With differentiation now in hand, you're in a position to see exactly what Madhava was doing, and why it turned out to matter so much to everyone who came after him.

The motivating problem is a practical one: how do you actually compute the value of a complicated function, like e raised to some power, by hand, using nothing but addition, multiplication, and division? There's no simple arithmetic shortcut for e^0.5. But there is a way to approximate it to an arbitrary degree of precision, by representing the function as an infinite sum of simpler polynomial terms, a Taylor series.

The function e^x has a genuinely convenient property for this: every one of its derivatives is e^x itself, so at a = 0, every term f(0), f'(0), f''(0), and so on, equals e^0 = 1. The series simplifies beautifully:

e^x ≈ 1 + x + x²/2! + x³/3! + x⁴/4! + ...

Try it yourself with just the first five terms, approximating e^0.5:

The true value of e^0.5 is approximately 1.6487. Five terms of simple arithmetic, addition, multiplication, division by small factorials, land within 0.0003 of the exact answer, and each additional term tightens the approximation further. This is precisely the kind of machinery Madhava had already built for sine and cosine centuries earlier, arrived at through geometric reasoning rather than derivatives, since he had no derivative concept to build from.

In Europe, the general method for building a series like this out of any function's derivatives was published by the English mathematician Brook Taylor in 1715, in a book called Methodus Incrementorum Directa et Inversa. It's worth noting Taylor wasn't first even within Europe, James Gregory, Newton, Leibniz, and Johann Bernoulli had each stumbled onto pieces of the same idea earlier, usually as one-off special cases rather than a general method. And despite being published, Taylor's general result sat largely unappreciated for nearly six decades, which is a strange thing to sit with, an idea this useful, ignored for that long.

It took until 1772 for Joseph-Louis Lagrange to recognize and publicly declare Taylor's expansion the fundamental principle underlying the whole of differential calculus, the moment the idea finally got the recognition its own discoverer never quite received in his lifetime. Between a mathematician in 14th century Kerala solving the same underlying problem centuries early with no successors to build on it, and a mathematician in 18th century England solving it generally and having to wait sixty years to be believed, it's a useful reminder that a mathematical idea's importance and its timing rarely arrive together.

When One Variable Isn't Enough: A First Taste of Multivariable Calculus

Everything built so far in this article involves a single changing quantity at a time, position depending on time, area depending on one curve. Most real quantities depend on more than one thing at once, and calculus had to stretch to cover that too.

Take the volume of a cylinder, V = π·r²·h, which depends on both its radius r and its height h. Ask a seemingly simple question: how does the volume change? The honest answer is that it depends entirely on which dimension you're changing, and calculus needs a way to isolate one variable's effect while temporarily freezing the other.

That curved ∂ symbol, rather than the ordinary d, signals a partial derivative, a rate of change with respect to just one variable, with every other variable held deliberately, temporarily still.

Put real numbers in to see how different the two answers actually are. Take a cylinder with radius 3 and height 5:

For this particular cylinder, nudging the radius up by a tiny amount increases the volume more than three times faster than nudging the height up by that same tiny amount. Stretching a cylinder taller and widening it have genuinely different effects on its volume, and a single-variable derivative has no way to even ask which one you mean. Multivariable calculus, built out through the 18th and 19th centuries by mathematicians including Euler and Lagrange, is exactly this idea generalized: extending every tool built so far, limits, derivatives, integrals, differential equations, to situations with more than one moving part at once, which describes almost everything in physics, economics, and engineering more honestly than a single variable ever could.

From a Clever Trick to a Competitive Sport: The Bernoullis and the Brachistochrone

Pick the historical thread back up here, in 1696, a decade after Leibniz published his notation, while the toolkit built up over the last several sections was still being worked out in scattered pieces across Europe. Calculus didn't stay a private technique for long. Within twenty years of Newton and Leibniz's work, it had become Europe's favorite intellectual blood sport, and nothing captures that better than what happened that June.

Johann Bernoulli, a brilliant and famously combative Swiss mathematician (and the same younger brother who'd outpaced Jacob on the catenary five years earlier), published an open challenge in Leibniz's journal, addressed to "the most brilliant mathematicians in the world": given two points, one higher than the other, find the shape of the curve along which a bead, sliding under gravity alone, would travel between them in the least possible time. Not a straight line, that's the shortest distance, not the fastest route. The problem became known as the brachistochrone, Greek for "shortest time."

Six mathematicians eventually solved it: Johann Bernoulli himself, Jacob Bernoulli, Leibniz, Guillaume de l'Hôpital, and, arriving anonymously through the post, one more. Isaac Newton, by then largely retired from mathematics and running the Royal Mint, received the challenge one afternoon after a long day's work. According to the account passed down by his family, he sat down that evening and had the full solution worked out by four o'clock the next morning, a curve called a cycloid, the path traced by a point on the rim of a rolling wheel. He mailed the solution to the Royal Society with no name attached.

Johann Bernoulli read it and knew immediately who'd written it. His reported response has outlived nearly everything else about the episode: tanquam ex ungue leonem, "we recognize the lion by his claw." Imagine being good enough at something that people can identify your handwriting from the shape of your reasoning alone. The problem also gave birth to an entirely new branch of mathematics on the spot, the calculus of variations, which doesn't ask for the derivative of a function, but for the function itself that makes some other quantity, like total travel time, as small (or large) as possible. Calculus had stopped being a tool for solving individual problems and started becoming a machine for generating entirely new kinds of problems.

Euler Turns a Bag of Tricks Into a Discipline

Calculus's inventors didn't get long to enjoy their invention before it curdled into a fight. In 1708, Newton's ally John Keill accused Leibniz outright of plagiarism in a paper for the Royal Society. Leibniz defended himself in print in 1711, citing his own private notes from 1675, written before he'd seen any of Newton's work. Newton himself escalated things in 1715 with an anonymous, hostile review of the dispute, effectively refereeing his own case.

The fight consumed the mathematical world for years and left a genuinely petty legacy: British mathematicians, out of loyalty to Newton, kept using his clumsier fluxion notation for another century, while Continental Europe adopted Leibniz's far cleaner system and pulled decisively ahead in mathematical progress. Most historians today agree both men developed calculus independently and legitimately.

For its first several decades, calculus also lived almost entirely in the hands of a small circle of geniuses who'd either corresponded directly with Newton or Leibniz, or studied under someone who had. There was no textbook you could hand a promising student. There wasn't even standard notation everyone agreed on, which means that if you'd been born in 1700 and wanted to learn calculus, you couldn't have. There was nothing to learn it from.

Leonhard Euler, working through the middle of the 18th century, squarely inside the Continental tradition the dispute above had just pulled ahead, fixed both problems, and did so with a level of output historians still find slightly unbelievable, he's credited with more surviving published pages than almost any mathematician in history, working through blindness in his later decades by dictating equations aloud to assistants.

Euler introduced the notation f(x) to describe a function, first in 1734 and then standardized in his 1748 book Introductio in Analysin Infinitorum. He introduced the summation symbol Σ, and was the first to use e for the base of the natural logarithm, the same e that shows up in both the growth-and-decay differential equation and the infinite series worked out above.

Then, across two systematic treatises, Institutiones Calculi Differentialis (1755) and the three-volume Institutiones Calculi Integralis (1768-1770), he laid out differentiation and integration as an organized, learnable discipline for the first time, complete with worked techniques, in something close to the order still used in textbooks today, including the one you're probably learning from right now.

Almost every piece of notation and organizational structure a modern calculus course actually uses, f(x), the systematic rules, the very idea of calculus as a subject with a beginning, middle, and end rather than a private toolkit, traces back to Euler's reorganization, not directly to Newton's fluxions or even Leibniz's original papers.

Ghosts of Departed Quantities: When Calculus Got Called Out

There was a problem sitting underneath all of this progress, and calculus's harshest, wittiest critic wasn't a mathematician at all, he was a bishop.

In 1734, the same year Euler introduced his function notation, George Berkeley, Bishop of Cloyne, published a pamphlet called The Analyst, aimed squarely at Newton's method of fluxions. Berkeley wasn't disputing that calculus worked, its predictions about planetary motion and falling bodies were plainly, repeatedly correct. His target was sharper, and honestly, fair: nobody could actually explain, with any logical rigor, why the method was valid. The whole enterprise leaned on a quantity that got treated as nonzero right up until the exact moment it became convenient to treat it as zero. If that sounds like a genuine loophole to you, you're in good company, it took mathematicians decades to properly answer it. Berkeley's own words landed like a slap:

"And what are these fluxions? The velocities of evanescent increments? And what are these same evanescent increments? They are neither finite quantities, nor quantities infinitely small, nor yet nothing. May we not call them ghosts of departed quantities?"

It was a genuinely difficult question to answer, and for the better part of a century, nobody quite managed to. Mathematicians like Colin Maclaurin tried patching the foundations in the years right after Berkeley's attack, with only partial success. Calculus kept being used, kept producing correct answers, kept getting built into physics and engineering, all while its own internal logic remained, by its critics' fair assessment, a little haunted.

Cauchy, Riemann, and Weierstrass Finally Close the Loop

The fix took nearly ninety years to arrive, and it came in stages.

Augustin-Louis Cauchy, in his 1821 textbook Cours d'Analyse, made the first serious attempt to define a limit properly, not as a mysterious vanishing quantity, but as a precise relationship between how close an input gets to a target and how close the output gets in response, essentially formalizing the "approaching a value" intuition you built earlier in this article. It was a real advance, but still left some gaps mathematicians would spend decades finding.

One of those gaps concerned the integral itself, the exact same rectangle-shrinking idea built up earlier. What functions could actually be integrated this way? Bernhard Riemann, working through his 1854 postdoctoral thesis under the mathematician Peter Gustav Lejeune Dirichlet, gave the rectangle-shrinking process its first fully rigorous definition, what's now called the Riemann integral.

Dirichlet had been pushing his students to think about genuinely strange functions, including one, sometimes called Dirichlet's function, that equals 1 at every rational number and 0 at every irrational number, a function so relentlessly discontinuous it can't be sketched at all, not even in principle. Riemann's formal definition needed to be precise enough to say clearly which functions, like this one, could not be integrated by the method, and which merely-strange ones could.

The final piece came from Karl Weierstrass, working through the 1850s to 1870s, who replaced the last remaining hand-wavy language, "as h approaches zero," "gets arbitrarily close", with the fully symbolic epsilon-delta definition of a limit still taught today: for any margin of error you name (traditionally called epsilon), a close enough input, within some distance delta of the target, is guaranteed to produce an output within that margin.

Made concrete, this reads less intimidating than it sounds. Take the simple limit lim (x→2) of 3x, which should equal 6. Someone skeptical is allowed to demand any margin of error they like, say epsilon = 0.03, and the claim is that a close enough delta always exists to guarantee it:

No infinitesimals required anywhere in that argument, no ghosts, just an ordinary number (delta) produced from another ordinary number (epsilon) by simple algebra. Ask for a stricter margin, say epsilon = 0.0003, and the same method just produces a smaller delta, 0.0001, every time, for any margin at all. This is what finally, completely closed the loop opened by Zeno of Elea roughly 2,300 years earlier: an infinite process genuinely can be shown, with total rigor, to converge to an exact value, and mathematics could now say precisely why, in a form Bishop Berkeley himself couldn't have picked apart.

The calculus you're learning in a classroom today isn't really Newton's calculus, or even Leibniz's. It's the twice-rebuilt version, patched by Cauchy, formalized by Riemann, and made fully airtight by Weierstrass, roughly a century and a half after the original argument between two exhausted geniuses in England and Germany.

How the Branches Actually Fit Together

Every idea built up across this article solves a genuinely different problem. Limits underpin differentiation and integration, which are each other's inverse, and together they feed differential equations and infinite series, both of which feed into the multivariable extension of everything.

Conclusion & Key Takeaways

Calculus wasn't handed down as a finished formula sheet, and if it ever felt that way in a classroom, that's a fair thing to be a little annoyed about. It was assembled, argued over, publicly humiliated, and rebuilt twice, by people trying to solve problems that had nothing to do with passing an exam, a philosopher questioning whether motion was even logically possible, a mathematician stacking infinitely many triangles to outsmart a curve, an astronomer in Kerala trying to build better sine tables three centuries before anyone in Europe needed the same trick, another astronomer annoyed at how wine merchants measured barrels, a plague-quarantined student not much older than you inventing physics and math at the same time because he needed both to finish either, a bishop who thought the whole thing was a bit of a con and was, for nearly a century, not entirely wrong, and a mathematician who lost his own public challenge to his younger brother but still gave the world the word "integral."

Every formula you'll be handed in a textbook is the compressed residue of one of those stories. You're allowed to ask which one, every time.

There's one last twist worth carrying with you past the last section. Newton and Leibniz's original, informal language of infinitesimals, quantities infinitely small but not quite zero, the exact thing Berkeley mocked as ghosts, had been treated as a convenient fiction for nearly a century after Weierstrass's fix, useful for building intuition but not considered logically real. In the 1960s, the logician Abraham Robinson proved that infinitesimals could, in fact, be built on completely solid logical ground after all, in a framework called non-standard analysis. The founders' original instinct, the one their own harshest critic tore apart in 1734, turned out to be recoverable after all. It just took two hundred and thirty years longer than expected, which might be the single best argument in this entire article for not being embarrassed when your first intuition about something turns out to need work.

Sources

Archimedes, Quadrature of the Parabola (c. 225 BC); Madhava of Sangamagrama, infinite series for sine, cosine, arctangent, and pi, as preserved in later Kerala school texts (c. 1400-1425); Johannes Kepler, Nova Stereometria Doliorum Vinariorum (1615) and Astronomia Nova (1609); Pierre de Fermat's correspondence on the method of adequality (1630s); Isaac Newton, unpublished fluxion manuscripts (1665-1666), published in part as an appendix to Opticks (1704) and posthumously in full (1736); Gottfried Wilhelm Leibniz, notation for differential and integral calculus, published in Acta Eruditorum (1684-1686); Jacob Bernoulli's catenary challenge and the origin of the term "integral," Acta Eruditorum (1690-1691); Johann Bernoulli's brachistochrone challenge and its solutions, Acta Eruditorum (1696-1697); Brook Taylor, Methodus Incrementorum Directa et Inversa (1715); Leonhard Euler, Introductio in Analysin Infinitorum (1748), Institutiones Calculi Differentialis (1755), Institutiones Calculi Integralis (1768-1770); George Berkeley, The Analyst (1734); Augustin-Louis Cauchy, Cours d'Analyse (1821); Bernhard Riemann, habilitation thesis on integrability (1854); Karl Weierstrass, rigorous formalization of the limit concept (1850s-1870s); Abraham Robinson, Non-standard Analysis (1966); MacTutor History of Mathematics Archive, University of St Andrews, "The Rise of Calculus" and "Madhava"