03Derivative & Integral

Chapter 2 handed you one tool and one habit, and this chapter spends both. The tool is the limit: a number, not a journey, and therefore a thing you may do algebra to. The habit is the order. Prove the thing exists, then compute with it. We closed by pointing that tool at a curve and slicing the area underneath into thin rectangles. That was a promise, and this chapter pays it in full. The ending first, because the ending is the whole reason to walk the road. Calculus is change under a microscope. Zoom far enough into any smooth curve and it becomes a straight line, and that line's slope is the derivative. Now add every one of those tiny straight bits back up and you recover the whole change, which is the integral. So they were never two ideas. They are one sentence read in opposite directions, and by the end you will have watched that sentence turn around. We'll start with a question that sounds simple and is quietly broken: what is the slope of a curve? We'll fix it with Chapter 2's limit and no cheating, and I will show you exactly where the famous 0/0 is supposed to be so you can see that it isn't there. Then we'll build the one rung every calculus course skips — the h² graveyard — and let it carry every rule in the chapter for free. Then we run the question backwards, and the two halves close on each other.

Here's where this sits. Part 0 is the gap zone and this chapter is its hinge, because almost everything above it in the course is one of these two ideas wearing different clothes. Chapter 4 stacks the higher derivatives into Taylor. Chapter 11 turns the area under a curve into probability, and ∫x·f(x)dx into expectation. Chapter 28 will call f' Delta and f'' Gamma and trade them for a living.

So let's go and get it. We start with a question you have almost certainly been given a wrong answer to.

01A curve does not have a slope

You already know what slope means for a straight line: it's rise over run, so you go along a bit, see how much you went up, and divide. One line, one number, and that number is the same everywhere along it. That's the whole story for lines, and school was right about it.

Now here is a stock's price over one morning. It wanders. Ask the natural question — how fast is it moving? — and watch what happens to the honest answer.

Pick two moments and read the price at each of them. The line joining those two points is a real line you can actually draw, and it has a real slope you can actually compute. So nothing exotic has happened yet. But drag one of the points and the number changes.

Sit with what you just felt, because it is the rung the whole chapter stands on. On a curve, the answer genuinely depends on which two points you picked, and that isn't you failing — that's the question being under-specified. So "the slope of the curve" is not a hard question with a hidden answer waiting at the end of it. As stated, it is a question with no answer at all, and nobody ever tells you that.

What we can compute has a name. Over the stretch from a to b, the number Δf/Δx is the average rate of change. That triangle is the Greek letter delta, and it just means "the change in". So Δf is the change in the price and Δx is the change in the time. The line through the two points is called a secant, from the Latin for cutting, because it cuts the curve at two places.

Notice that we have not cheated once. This is still rise over run, on a line we drew, using a function that hands us a price for every time — which is Chapter 1's machine, unchanged. All we did was refuse to answer a question we cannot answer yet.

02Shrink the gap, and nothing breaks

The secant is honest but ambiguous, so let's remove the ambiguity the only way available to us: stop moving one of the points. Nail the left point down at x, and put the right point at x + h, where h is nothing more exotic than a name for the gap between them. Now the secant slope is

[ f(x+h) − f(x) ] / h

and it has a name: the difference quotient. One point is fixed, so there's only one thing left to vary. Shrink h.

Here is the single most famous tear in all of calculus, and the whole thing happens in about one inch of page. Set h = 0 and the quotient reads 0/0, which is not a number and not anything else either. Leave h ≠ 0 and you still have a secant, which is not the thing you wanted. Every reader smells a cheat here, and most books sprint past it with a mumbled "h approaches zero but never equals zero." That sentence is technically true and it teaches nothing, and readers who never get this straight spend years quietly believing that calculus rests on a fib.

So let's do it slowly, and let's do the numbers before the algebra. Take f(x) = x² and stand at x = 1. Compute the secant slope for a few gaps and call the answer out loud before I say anything.

At h = 0.1 the slope is 2.1, at h = 0.01 it's 2.01, and at h = 0.001 it's 2.001. You said 2, and you said it out loud before anything clever happened to you. The number is simply there, sitting in a table you can check by hand with a calculator. That's Chapter 2's limit, running on ordinary arithmetic.

Now the algebra, in that order, and watch every step. Expand the top: (1+h)² − 1 = 1 + 2h + h² − 1 = 2h + h². Divide by h: you get 2 + h. Stop there for a second, because that cancellation is the whole ballgame. It is 100% legal, because h ≠ 0 the entire time we are doing it. We never divided by zero. Not once.

Then, and only then, we take the limit — of 2 + h, which is an ordinary expression with no division in it at all. Chapter 2 tells us what that limit is without breaking a sweat: it's 2. So the move is simplify while h is alive, then take the limit. The 0/0 everyone warns you about never appears, because we never evaluated the quotient at zero. We simplified it first.

That number has a name: it is the derivative of f at 1, it is written f'(1), and it equals 2. Here is the definition in full, and every piece of it is something you just built by hand:

f'(a) = limh→0 [ f(a+h) − f(a) ] / h

Now, what is that number, geometrically? We can already compute it, but I want you to see it, so take a smooth curve, put your eye on a single point, and magnify. Not a jump cut — one continuous zoom, and keep going.

The curve flattens, and if you push the magnification far enough the pixels become a straight line and the printed slope stops changing. That's it — that's the derivative, seen instead of computed. The property has a name, local linearity, and it is exactly what the word smooth was always quietly pointing at. Smooth means locally straight, if you look close enough.

Which lets us define the tangent line properly, and the definition is simply that it is the line you get by zooming. You may have been told a tangent is "the line that touches the curve at one point," and I want to be blunt about that: it is both false and useless. A tangent can cross the curve somewhere else entirely and still be the tangent. Touching was never the point. The zoom was.

Now the notation. I owe you a debt here, and I'd rather settle it up front than leave it lying around.

There are two names for this one object. f'(x) is Lagrange's notation, and it's just a tick mark. dy/dx is Leibniz's, and it causes trouble, because it is drawn as a fraction while we just defined it as a limit. So which is it? Here is the honest answer, which almost nobody gives you. dy/dx is one symbol, meaning that limit. It is not a fraction, and you cannot go slicing the dx off the bottom because you feel like it.

But that's only half of it, and the other half is why the confusion never dies. dy/dx was built out of Δy/Δx, and Δy/Δx really was a fraction — an honest ratio of two ordinary numbers. That ancestry is exactly why dy/dx will behave like a fraction when we get to the chain rule. We are going to cash that in later, with an actual argument, not a wink. I'm flagging it now so you don't spend the next hour wondering whether you're allowed to cancel.

03The promotion nobody announces

We have a number at a point. That's a small thing to own, and re-running the algebra at every point we care about would be a miserable way to live. So stop feeding it numbers, and watch what falls out.

Run the exact same limit at x = 1, then again at x = 2, then again at x = 3, and you get 2, then 4, then 6. It is the same seven lines of algebra every single time, and only one thing ever changed.

Now run it a fourth time with the letter x sitting in the slot where the number sat. The machine never changed. Only what we fed it. And out falls f'(x) = 2x — which is not a number. It's a function: a machine that hands you the slope at whatever point you feed it.

Stop here, because a rung just got skipped in most books and it quietly wrecks people. On one line they compute f'(2) = 4, and on the very next line they write f'(x) = 2x. Nobody flags that x just got promoted — from "a specific place we're standing" to "any place at all." The reader nods, and then cannot answer the simplest question in the topic.

So let's answer it, out loud, right now. The word derivative wears two hats. f'(2) is a number. f' is a function. Same word, two different kinds of object, and the switch between them happens in silence one line apart. Whenever you feel lost in a calculus argument, check which hat is on.

Now let's cross-check the new function against the old one, because a claim this big should be checkable. Plot f and f' one above the other on a shared time axis and read the story off both.

Where f is climbing, f' sits above zero. Where f is falling, f' sits below zero. At f's peak, f' crosses zero — the curve is momentarily flat up there, so its slope is momentarily nothing. Two graphs, one story. You can read either one from the other, and that is the first sign that f' is a real object and not bookkeeping.

04★ The h² graveyard

Now a rung almost nobody builds, and everything after it leans on this one.

You have already seen it happen. We wrote (2h + h²)/h = 2 + h and let h go to zero. Fine. But you are about to see "+ h² … which vanishes" in a product-rule proof, and again in the Fundamental Theorem's error term, and again in Chapter 4's Taylor remainder. Every single time it will feel like the author waving away something inconvenient. "Small enough to ignore" is an engineer's shrug, and a shrug is not a proof, so readers stop trusting the derivations. Fair enough. Let's kill that suspicion permanently.

The precise confusion is that people don't distinguish small from second-order small. Take the smallest possible example. Draw two bars, one of length h and one of length h², and drive h down from 1.

At h = 1 the two bars are exactly equal, and the graveyard doesn't exist yet. At h = 0.1 the h² bar is a tenth of the h bar, and by h = 0.001 it is a thousandth of it and has vanished off the screen entirely.

Here's the only line that matters, and I want it flat and plain: h²/h = h → 0. Put a number through it: at h = 0.001 the square is 0.000001, and dividing by 0.001 leaves 0.001 again. We are not ignoring the corner, because we divide by h first and only then look at what survives. What's left of h² after that division is h itself, and h goes to zero honestly by Chapter 2's definition, with no shrug anywhere in sight. Nothing is approximate and nothing is dropped for convenience. It genuinely goes to zero after the division.

So here is the vocabulary, and it will do real work for the rest of the course. A term of order h is first-order, and it survives the division and matters. A term of order h² or worse is second-order, and in the limit it is exactly nothing. That asymmetry — not any trick — is what makes every rule in this chapter work.

And now let me plant a fuse, because this course is going to need it. Everything in calculus that ever feels like a cheat is this one fact. Every one. So remember the graveyard, and remember that we buried things here on purpose. In Chapter 26 we will meet a wiggle so violent that its square does not die — a path where dW² = dt instead of vanishing. The whole of modern finance falls out of that one exception. Somebody is going to climb out of this graveyard, and when they do, you'll want to have watched the burial.

05Four rules, none of them decreed

Time to make the machine cheap to run. And every rule below gets derived, because a rule you can re-derive is a rule you can never lose.

Start with the most-memorized, least-understood spell in mathematics: "bring the power down and knock one off." Where does that come from? If you don't know, you can't tell whether it works for n = ½, or n = −1, and you'll be guessing forever. So let's find the n.

We did x² already, so let's do x³ by hand and expand (x+h)³ = x³ + 3x²h + 3xh² + h³. Subtract x³, divide by h, and what you're left holding is 3x² + 3xh + h². Those last two terms still carry an h, so they're graveyard-bound, and the answer is 3x². Now predict: what will x⁴ give?

You said 4x³ from the pattern, and you were right. But pattern-spotting isn't understanding, so here's why. Look inside (x+h)ⁿ: it's n brackets standing in a row, and you pick one item out of each of them. Exactly one kind of term ends up with a single h in it, and that's the term where you took the h from one bracket and the x from all the others. How many ways can that happen? Chapter 1 already answered that: choose which one bracket donates, which is C(n,1) = n. Take n = 4: four brackets, so C(4,1) = 4 ways to pick the donor, and that 4 is the 4 sitting in front of 4x³.

That's the whole formula. d/dx xⁿ = n·xⁿ⁻¹. Feed it n = 3 and out comes 3x², the answer we ground out by hand. Every other h-carrying term picked up h² or worse and is already buried. So the n in n·xⁿ⁻¹ was a count the entire time. It is not a spell. It is Chapter 1's combinations, showing up exactly where you'd expect them if you'd expanded the bracket.

Two more rules fall out in one line each, and they're worth stating only because they're so easy that people assume they're deep. Derivatives add: (f + g)' = f' + g', because the difference quotient of a sum is the sum of the difference quotients. And constants ride along outside: (c·f)' = c·f', because you can factor c out of the quotient before taking the limit at all. Check both at x = 2: x² + x³ has slope 4 + 12 = 16, and 5x² has slope 20. Done.

Now the one your gut is going to get wrong, and I mean dead wrong. What is (fg)'? Almost everyone reaches for f'g', because sums behaved that way ten seconds ago and the mind generalises. Let's test it on the smallest example there is: x · x.

The gut's rule says f'g' = 1 · 1 = 1. But x · x is x², and we derived that ourselves: its slope is 2x. So the gut's rule claims the slope of x² is 1 everywhere. That's not slightly off. That's dead. Good. Now that the wrong belief is out of the room, here's the picture that shows you the right one.

Picture a rectangle with sides f and g, whose area is therefore fg. Now grow both of its sides a little, f by Δf and g by Δg, and the new area arrives in three pieces you can literally see. There's a strip along the bottom of area g·Δf, and a strip up the side of area f·Δg. And then there's a little corner of area Δf·Δg.

Divide the whole change by h, and the two strips hand you f'g and fg'. The corner is a product of two small things, so it's second-order, so it's already in the graveyard — and it is the only casualty in the whole picture. That's the entire rule:

(fg)' = f'g + fg'

And now you can see why the answer has two terms instead of one: a rectangle has two sides to grow. On x · x it hands back 1·x + x·1 = 2x. Notice we never added and subtracted a mysterious f(x+h)g(x) out of nowhere — that's the standard proof, and it verifies the rule while explaining nothing at all.

Next: what if one thing depends on another, which depends on a third? Before any calculus, answer this. A big gear turns 3 times for every turn of your hand. A small gear turns 5 times for every turn of the big one. How fast does the small gear turn per turn of your hand?

You said 15, instantly, and nobody had to teach you any calculus to get it. Your gut already owns the chain rule — it just doesn't know that it does. Rates multiply through a linkage, and that's the whole idea, and you've believed it since you were a child.

So let's ground it. If z depends on y and y depends on x, then Δz/Δx = (Δz/Δy)·(Δy/Δx). That is the gear train in symbols: 5 × 3 = 15. The Δy on the top of one factor and the bottom of the other is a number, and it cancels, and that is boring, ordinary, legal algebra. It is legal precisely because before the limit these really are fractions — which is the debt I flagged earlier, now paid. Only after cancelling do we let h → 0 on both factors, and each factor is a difference quotient in its own right. So:

dz/dx = (dz/dy) · (dy/dx)

So dy/dx is not a fraction, but it is the limit of one, and that is exactly why it acts like one here. No wink required. One piece of fine print, stated rather than hidden: if Δy happens to be zero the cancellation needs care. The standard fix is routine and we're not going to do it here. I'd rather flag it than pretend it isn't there.

And that leaves one rule you were probably dreading, so let me take it off the table. There is no fourth rule.

f/g is just f · g⁻¹, so it's the product rule and the chain rule shaking hands, and the quotient rule falls out of the two of them with no new ideas at all. Derived, not decreed. If you ever forget its shape you can rebuild it in about thirty seconds, which is a far better place to stand than remembering it.

One last move before we turn around, and it costs nothing: f' is a function, so you can feed it straight back into the machine again. Do that and you get f'', the second derivative — the rate at which the rate itself is changing. Mechanically it's trivial. Semantically it is slippery, and the slipperiness is where money gets lost.

Here's the concrete version, and you have felt this one in your body: position, then speed, then acceleration. Brake in a car. Your speed is still positive, because you're going forward, while your acceleration is negative at the same moment. That's f' > 0 together with f'' < 0, and it is completely ordinary.

Now predict, on something that will matter for the rest of this course. Two price paths end at the same point. One has f'' > 0 the whole way, the other has f'' < 0. Which holder slept better?

The gut wants to collapse f'' > 0 into "going up". It does not mean that. A stock can be falling, with f' < 0, and still have f'' > 0 — still losing money, but losing it more slowly each minute, with the curve bending toward the floor. All four sign combinations are real, and your gut only believes in two of them.

So hold it flat and plain: f' tells you where you're going, and f'' tells you how the story is bending. That bend is called curvature. Chapter 4 will rebuild an entire function out of this stack of derivatives, and Chapter 28 will call f'' Gamma and put a price on it.

06Backwards: given the rate, how much piled up?

This whole chapter has asked one question, over and over: given the amount, how fast is it changing? Now run it backwards. Given the rate, how much piled up?

And here I have to fix the single biggest unearned assertion in every calculus course. You are about to be told that the integral is "the area under the curve." Area gets announced as the goal, and you accept it the way you accept a rule of a board game — arbitrary, but fine. That's how integration becomes a second, unrelated subject that happens to share a book with derivatives. And then the theorem that joins them feels like a coincidence between two strangers.

It isn't a coincidence at all, and here's the sentence that's missing from almost every telling of it. Nobody does the units.

Drive at a constant 60 miles per hour for 2 hours, and the distance is 60 × 2 = 120 miles. No calculus required, and you did it in your head without noticing. But look hard at what you just multiplied: a height by a width. On the speed graph that's a flat line at 60, and the thing you computed is literally the rectangle underneath it.

Say the units out loud, because they're the whole argument: (miles/hour) · (hours) = miles. Ours read 60 miles/hour × 2 hours = 120 miles. The hours cancel, and that cancellation is the entire reason area shows up here at all. So area was never the goal — it is the bookkeeping of a multiplication. Rate times time is a rectangle, and it always was.

Now change speed three times along the way. You get three rectangles, you add them up, you're done. It's still exact, still just arithmetic, and the speed graph has become a staircase. That's the piecewise-constant case, and it needs nothing we don't already own.

Now let the speed vary smoothly, and the trick breaks in your hands. There are no rectangles left at all, because the height is different at every single instant. Sit in that gap for a second before we fill it, because that gap is the honest problem. It is not "find the area." It's "the multiplication that worked twice just stopped working."

So fix it the same way we fixed slope: chop the varying rate into n slices so thin that the rate is almost constant across each one. On every slice, multiply rate by width to get a rectangle that is legal and finite, then add every one of them up. Then let n → ∞.

Your gut is about to revolt, so let's pre-empt it. Infinitely many pieces, each one infinitely small, sounds like 0 × ∞, and that really is nonsense. But Chapter 2 already did the scary part for us, because infinitely many shrinking terms summing to a finite number is not a paradox at all. It's the geometric series, and you held one in your hands. Same species. The fear is already spent.

So let's run it on the smallest example whose answer you can check with a ruler: f(x) = x on the stretch from 0 to 1.

With n rectangles the sum is (1/n²)(1 + 2 + … + n). That inner sum is Chapter 2's arithmetic series, n(n+1)/2, so the whole thing collapses to (n+1)/(2n). Four rectangles give 5/8, or 0.625. A hundred give 101/200, or 0.505. Let n → ∞ and it goes to 1/2. Which is the area of a triangle with base 1 and height 1 — a number you knew before you could read. The machine returned an answer you can verify independently, and that receipt buys a lot of trust.

That sum has a name: the Riemann sum, written Σ f(xᵢ)Δx. Its limit is the definite integral, and it gets a symbol:

∫ab f(x) dx

Now let me translate those hieroglyphs, because nobody ever bothers and they are both completely mundane. ∫ is a stretched letter S, and it stands for Sum. Leibniz literally drew a long S. And dx is what's left of Δx after the shrink — it's the width of one sliver. So read the symbol out loud: sum up, over all the slivers, height times width. It was a shopping list the whole time.

07★ The wall you push — the keystone

Everything so far has been setup. This is the section the whole chapter exists for.

Here's the exact inch where the Fundamental Theorem gets lost, and it is not in the proof. It's one step earlier. So far you have only ever met the integral as a number: from 2 to 5, computed, done. So when a book writes d/dx ∫ₐˣ f(t)dt, it is asking you to differentiate a number. Numbers don't have derivatives. The theorem is nonsense from its first symbol, and no amount of rigour repairs that, because the missing thing isn't rigour. It's the object.

So let's build the object physically before we touch it. A tank with a tap. f(x) is the flow rate right now. A(x) is how much water is in the tank. Slide time forward and watch both at once: the tap's dial on top, the tank's level below.

That's all we did. We took the definite integral and replaced the fixed right-hand edge b with a variable x. One letter. And now it's a function of where the wall currently stands:

A(x) = ∫ax f(t) dt

Kill the notation tear before it stalls you, because two letters just showed up and it looks like a typo. Inside that expression, t is a dummy variable. It's the name for the slivers we walk across between a and x, and it never escapes the integral. The x is the wall. Two different jobs, so they get two different letters. That's all it ever was.

Now the one question nobody thinks to ask, and it's available to us only because A is a function and functions have derivatives. How fast does the area grow?

Build the guess with your own hands before we prove anything. Set the tap to a constant 3. What is A(x)? The water piles up at a steady rate, so the area under the dial is just a rectangle, which makes A(x) = 3x. Now ask for its slope, and every rule you own hands you 3 — which is exactly the tap setting you dialled in. Coincidence?

Try another one, and set the tap to f(x) = x. The area under that dial is a triangle, so A(x) = x²/2, and at x = 1 that hands back the 1/2 you just checked against a ruler. Its derivative, by the power rule you derived yourself, is x — the tap setting again, and that's two for two. So commit out loud before the reveal: is A' = f always?

Now the stroke that proves it, and it takes exactly one push. Grab the right-hand wall and shove it right by a hair h, then freeze the frame and look hard at what you just added. It's a sliver — one thin rectangle, of width h and height f(x) — so its area is f(x)·h. Which gives:

A(x+h) − A(x) = f(x)·h + (a scrap along the top)

That scrap is the bit where f wiggles a little as it crosses the sliver's width. It's smaller than h times the wiggle, so it's second-order, which means it's already lying in the graveyard we dug. Divide the whole line by h:

[ A(x+h) − A(x) ] / h = f(x) + (something → 0)

Look hard at that left-hand side, because it's the difference quotient from the very start of this chapter, unchanged, the same one we've used all the way through. Take the limit. And there it is:

A'(x) = f(x)

That is the Fundamental Theorem of Calculus, part one, and no new machinery entered the room to get it. Not one piece. Now read what it says in plain words: the area accumulates at exactly the rate of the height at the edge. The tap's flow rate is the derivative of the tank's contents, and of course it is. The reason the area grows at rate f(x) is that you are pouring at rate f(x). It could not have been anything else.

So sit with the whole chapter for a second, because it just collapsed into one line. It was one sentence read in two directions the entire time. Read left to right, it's differentiation. Read right to left, it's integration. They were never two subjects, and that is why one undoes the other.

One small rung left before we can compute anything with it, and it looks so obvious that it usually gets assumed in silence. If a function's derivative is zero everywhere, what is the function?

Picture the tangent line riding along the graph. If it is dead flat at every single point, where could the curve possibly go? Suppose it got higher at b than at a: then somewhere between them it had to climb, and climbing means a positive slope somewhere, which we just forbade. So it can't. Zero rate of change means it never changes, and the function is constant.

Now the honest note, because "obvious" isn't always free. A derivative at a point is local information. Using it to conclude something about the function across a whole interval is a genuine step, and it has a real name: the Mean Value Theorem. Here's all it says, demystified. Over any stretch, the curve's average rate is achieved by its instantaneous rate at some point inside. So if the average is forced to be zero, the function ended where it started. The full proof lives in real analysis, we are not doing it here, and I'm not going to pretend we did.

08Subtraction instead of a million rectangles

Time to collect, and the payoff is enormous: you never have to sum a million rectangles again.

But first let's name the word-collision that poisons this whole topic. The symbol ∫ wears two hats. With limits on it, ∫ₐᵇ f is a definite integral — a number, the limit of those sums. Without limits, ∫ f dx is an indefinite integral — a family of functions. Same symbol, two completely different objects, and almost nobody says this out loud.

Here's the payoff itself, and it is absurdly cheap. Find any function F whose derivative is f — call it an antiderivative — and then

∫ab f = F(b) − F(a)

Areas, computed by subtraction. But that word "any" should bother you, and I want it to keep bothering you for a moment. There are infinitely many antiderivatives. They differ by a constant, the famous +C. So surely the answer depends on which one I grab?

Concrete first, before I answer a question you haven't properly asked yet. You found F(x) = x²/2. I found F(x) = x²/2 + 7. We can't both be right. Squirm for a second, then resolve it yourself with the rung we just built.

Our two answers have the same derivative, so their difference has derivative zero everywhere, and the rung we just built says a thing like that is a constant. That's not a coincidence. It's forced.

Now the arithmetic, and it's three lines. A is an antiderivative that happens to start at zero, so A(a) = 0. Any other F is A + C. Therefore F(b) − F(a) = (A(b) + C) − (A(a) + C) = A(b) − 0, which is the area. Run my x²/2 + 7 from 0 to 1: (1/2 + 7) − (0 + 7) = 1/2, the same triangle. The C cancels because it's the same C in both terms. You subtract the same starting offset you added. So it doesn't matter which antiderivative you grab, because subtraction forgets where you started counting from.

Which means integrating is just differentiating in reverse. Read every rule we derived right-to-left and you can compute real areas by hand.

∫ xⁿ dx = xn+1/(n+1) + C. Take n = 2: it gives x³/3, and differentiating x³/3 returns 3x²/3 = x². And the reason is not a spell: differentiate the right-hand side and the power rule hands you back exactly xⁿ. That check is free and always available, which makes guessing a legitimate method rather than a confession. Integration is the only branch of maths that grades its own homework.

Notice the one case that rule kills. At n = −1 you'd be dividing by zero. That hole exists, it's real, and the function that patches it is waiting in Chapter 5.

Now the honest part, the thing most courses never say — and it changes how you'll work. You have spent an hour learning a procedure for differentiating, so you expect a matching procedure for integrating. There isn't one. Differentiation is a machine, and it always terminates. Integration is a search that might simply fail.

Worse — and better, because knowing this is power. Most functions have no elementary antiderivative at all. Take e^(−x²), the bell curve we will need in Chapter 12 and again in Chapter 27, and it has none. Not because we aren't clever enough. Because it does not exist. That is exactly why the Normal CDF gets its own symbol N(x), why quants compute it numerically, and why half of this course's integrals will end their lives inside a Monte-Carlo loop. Told nothing about this, people assume their own incompetence the moment the search fails on them. Told plainly, you reach for a computer by judgement instead of by defeat. Knowing which side of that line you're standing on is the actual skill.

The whole chapter fits in one picture.

Look at what you're walking out with. You can point at any quantity and ask it both questions — how fast is it moving? and how much has piled up? — and you know, in your hands, why those two questions answer each other. The rate is the derivative of the total. That's the entire content of the sliver.

More importantly, you can re-derive rather than recall. The power rule comes from the binomial expansion, and the product rule comes from a growing rectangle. The chain rule comes from gears, and the Fundamental Theorem comes from pushing a wall. None of it is stored, and all of it is reachable. And you know the exact reason nothing was ever a cheat: h² dies after you divide by h.

Which leaves the fuse. In Chapter 26 you'll meet a path so jagged that dW² = dt refuses to die, and Itô's lemma is nothing but this chapter with one extra term left standing. That's not a new subject either. It's this one, with a survivor.

And the next rung is already under your feet. We just built f', then fed it back into the machine to get f'', and nothing stopped us from doing it again. So stack them all. It turns out that the whole function can be rebuilt, locally, out of that stack of derivatives at a single point — which is astonishing the first time you see it, and which is Taylor series. Its second term carries f'', and the sign of that term is convexity. That's Chapter 4.