"Gets close to" is motion language, and there's no motion here. A function is a static object; x isn't sliding anywhere. For 190 years everyone waved at this and it mostly worked — until it didn't, and mathematicians found functions where intuition gave flatly wrong answers.

Weierstrass's fix in the 1860s replaces motion with a challenge-and-response game:

Read it adversarially. Someone challenges you: "get f within 0.001 of L." You must produce a radius around a small enough to guarantee it. If you can meet every challenge, the limit exists.

The load-bearing detail is the leading 0 <. It explicitly excludes x = a. That single symbol is what makes limits able to talk about 0/0 at all.

One level deeper: this only works because the reals are complete — they have no gaps. A sequence of fractions can march toward √2, which is not a fraction. The real numbers were constructed (Dedekind, 1872) precisely so that anything that looks like it's approaching something actually has something to approach.

Completeness is the hidden foundation under the three big existence theorems of the course — Intermediate Value, Extreme Value, and Mean Value. All three are false over the rationals alone. BC states all three and proves none, so this sentence is the one the textbook is missing.

It means the form alone doesn't determine the answer. Consider three limits as x→0, all of the form 0/0:

Same form, three different answers. The form tells you nothing; you have to dig into the specific functions. Contrast 5/0, which is not indeterminate — it always blows up. Knowing the difference tells you whether there's work to do.

Apply the Factor Theorem to both halves of the fraction. If substituting x = a gives 0/0, then the numerator is zero at a and the denominator is zero at a. So (x − a) divides both. It's there by guarantee, not by luck. Your only job is to find it.

That's why "cancel and re-substitute" always works for rational functions, and why it's the first thing to try.

These are the limits that can't be done by algebra, because the functions involved aren't polynomials and there's no factor to cancel. Each one has to be established once, from a picture or a definition, and then it becomes a tool. Every trig and exponential derivative in the course rests on one of them.

Every step of that derivation used sector area = ½r²θ, which is only true in radians. In degrees the sector area picks up a factor of π/180, and the limit comes out to π/180 ≈ 0.01745 instead of 1.

Consequence: in degrees, d/dx[sin x] = (π/180)·cos x, and that ugly constant would propagate through every trig formula in mathematics forever. Radians are the unit that makes calculus clean — that's the entire reason they exist.

This limit is the derivative of e^x at x = 0 — write out the difference quotient with a = 0 and you get exactly this expression.

So the question "why is it 1?" is really the question "what is e?" And the answer: e is defined as the base for which the exponential curve has slope exactly 1 where it crosses the y-axis. Every exponential y = b^x passes through (0,1). They differ in how steeply. For b = 2 the slope there is about 0.693; for b = 3 it's about 1.099. Somewhere between 2 and 3 there's a base where the slope is exactly 1. That base is e ≈ 2.71828.

Everything convenient about e follows from that one choice — including d/dx[e^x] = e^x, the only function that is its own derivative.

Invest $1 at 100% annual interest. Compounded once: $2. Compounded twice a year at 50% each: (1 + ½)² = $2.25. Quarterly: (1 + ¼)⁴ ≈ $2.44. Daily: ≈ $2.7146. Continuously — the limit as n→∞ — you get e ≈ 2.71828, and no more. Compounding infinitely often doesn't give you infinite money.

Jacob Bernoulli found this in 1683 studying exactly this question. The general form with x in place of 1 is the same statement at interest rate x.

This matters later: it's why continuous growth models use e^kt, and it reappears in Unit 7 as the solution to y′ = ky.

ln x ≪ x^p: logarithms grow like the number of digits. Going from a thousand to a trillion multiplies x by a billion and raises ln x by about 21. Nothing polynomial can be beaten that slowly.

x^p ≪ e^x: differentiating a polynomial lowers the degree; do it p+1 times and it's gone. Differentiating e^x changes nothing. Apply L'Hôpital enough times and the polynomial dies while the exponential stands there.

e^x ≪ x!: e^x multiplies by a fixed e at each step; x! multiplies by an ever-growing factor. Fixed ratio always loses to growing ratio.

x! ≪ x^x: x! is x·(x−1)·(x−2)···, all factors below x; x^x is x multiplied by itself x times.

It's not a limit trick; it's the ranking of how fast things can grow, and it recurs constantly:

• In this course, Unit 10: whether an infinite series converges is entirely a question of whether its terms shrink fast enough. Every convergence test is the hierarchy in disguise.
• Improper integrals: ∫₁^∞ dx/x^p converges only when p > 1. The threshold is a growth-rate threshold.
• Computer science: an algorithm that takes ln n steps is fine at any scale; n² is usable; 2ⁿ is unusable past about n = 50; n! is unusable past about n = 15. Same ordering, and it's the reason some problems are considered intractable.
• Compound growth generally: exponential eventually beats polynomial always, however unfavourable the constants. A quantity growing 1% a year overtakes any fixed-power trend, given enough time. That's the mathematical content of most arguments about compounding.

Near the point a, each function is well approximated by its tangent line. If f(a) = g(a) = 0, those tangent lines both pass through zero there, so near a:

The ratio is then f′(a)(x−a) / g′(a)(x−a), and the (x−a) cancels. What's left is f′(a)/g′(a).

L'Hôpital is just cancelling the common factor again — the same move as every algebraic technique above, done with tangent lines instead of factoring. That's why it needs the 0/0 condition: without it, the tangent lines don't both pass through zero and the cancellation isn't available.

Historical aside worth having: L'Hôpital didn't discover it. He paid Johann Bernoulli a retainer for exclusive rights to his mathematical output, published it in the first-ever calculus textbook in 1696, and credited him only vaguely. Bernoulli complained about it for the rest of his life.

Continuity means taking the limit and evaluating the function commute — you can do them in either order and get the same thing. That's the useful form, because it's exactly what licenses "just substitute" as a limit technique, and it's what lets you pass a limit inside a continuous function later on.

Informally: a continuous curve can't get from below a line to above it without crossing it. Obvious — for the reals.

It is false over the rationals. Take f(x) = x² − 2 on [1, 2] with only rational inputs allowed. f(1) = −1, f(2) = 2, so it must cross zero — but the crossing point is √2, which isn't rational. The function jumps over zero without ever landing on it.

So IVT isn't a fact about continuity alone. It's a fact about continuity plus the completeness of the reals. This is the payoff of that earlier note about gaps, and it's the reason existence theorems in calculus are theorems rather than observations.

Weierstrass's function (1872): continuous everywhere, differentiable nowhere. A curve with no smooth point anywhere — no tangent line at any location, infinitely wrinkled at every scale.

Before it, mathematicians assumed continuity basically implied smoothness apart from isolated corners. Hermite called it "a lamentable plague." It's the ancestor of fractals and of Brownian motion — actual stock price paths and actual pollen-grain trajectories are exactly this kind of object. The intuition "continuous means you can draw it without lifting the pen, so it must have a direction almost everywhere" is false, and this is the counterexample.

Next stop — real analysis. Redo the entire year with proofs. Where ε-δ is the point, where completeness gets stated honestly, and where the pathological examples (Weierstrass's function, the Cantor set) stop being curiosities and become the objects that force the definitions.

Then — measure theory and the Lebesgue integral, which fixes the Riemann integral's limitations and is the actual foundation of modern probability. Also topology, which is what you get when you keep "nearness" and throw away distance and number entirely.

And the wildcard: nonstandard analysis (Robinson, 1966), which vindicates Leibniz by constructing infinitesimals rigorously. If she ever objects that treating dy/dx as a fraction "shouldn't be allowed," this is the field that says she was right the whole time.

The limit machinery is built. Now it gets used for the thing it was built for, and the course starts paying.

This unit is the highest-leverage one in BC. Everything downstream — optimization, integration by substitution, differential equations, Taylor series — is either an application of these rules or a reversal of them. Fluency here is worth more than fluency anywhere else, and gaps here compound for the rest of the year.

Two problems with it. First, it's nearly circular: you can't define the tangent line to a general curve without already having derivatives. (Circles are the exception — tangent means perpendicular to the radius — which is why the Greeks could do tangents to conics and nothing else for two thousand years.) Second, it doesn't generalize; "slope" stops meaning anything in higher dimensions.

Here's the alternative:

In words: the derivative is the multiplier in the best linear approximation to f near a. Zoom in far enough on a differentiable function and it becomes indistinguishable from a straight line. f′(a) is that line's slope, and E is the error you're making.

Why bother: this makes the error term explicit and nameable, and the error term is what the rest of the course is about. Linear approximation (Unit 4), Newton's method (Unit 4), Euler's method (Unit 7), and Taylor series (Unit 10) are all the same question — how good is this, and what's the next correction? Carrying this definition from September makes April easy.

Newton had calculus (he called derivatives "fluxions") by 1666, during the plague years, and didn't publish. Leibniz developed it independently and published in 1684. The resulting priority war was vicious and lasted decades.

Britain sided with Newton and kept his dot notation. The continent used Leibniz's dy/dx and ∫ — notation that suggests the correct manipulations. dy/dx looks like a fraction and behaves like one under the chain rule; ∫ f dx reads as "sum of height times width." British mathematics fell roughly a century behind as a result.

It's the strongest argument in the history of the subject that notation is not cosmetic. Interface design is leverage.

If f′(a) exists, the numerator [f(a+h) − f(a)] must be heading to 0 — otherwise, divided by a vanishing h, the quotient would blow up rather than converge. Numerator → 0 means f(a+h) → f(a), which is continuity.

The converse fails because a function can be perfectly connected and still have no well-defined direction at a point. |x| at 0 is the standard case: the secant slopes approach −1 from the left and +1 from the right. Both one-sided derivatives exist; they disagree; so the derivative doesn't exist.

Use the linear-approximation definition. Near a point, g magnifies a small input change by the factor g′. The result then enters f, which magnifies its input change by f′ — evaluated at g's output, which is g(a), not a. Two magnifications in sequence multiply.

In Leibniz notation the du's appear to cancel: (dy/du)(du/dx) = dy/dx. That's a genuinely good intuition and it's why Leibniz's notation won. The naive proof — multiply and divide by Δu — breaks if Δu happens to be 0, and patching that is why textbooks get evasive here. The patch is routine; the intuition is correct.

The chain rule is roughly a third of the course wearing different hats. Implicit differentiation is the chain rule with y treated as a function of x. Related rates is the chain rule with t as the hidden variable. u-substitution in Unit 6 is the chain rule run backwards. Each is taught as a separate topic; they're one rule.

Trig is where rust shows worst, and it's load-bearing for the rest of the year — every integration technique in Unit 6, all of Unit 9 (polar), and the Taylor series in Unit 10 rest on it. This section is the whole of what BC actually uses.

A point moving around the unit circle at unit speed is at (cos t, sin t). Its velocity vector is tangent to the circle, same length, rotated 90° counterclockwise from the position vector. Rotating (cos t, sin t) by 90° gives (−sin t, cos t).

Read off the components: the x-component says d/dt[cos t] = −sin t, and the y-component says d/dt[sin t] = cos t. Both derivatives at once, no limits, no identities — geometry did the work.

This is also the seed of e^it = cos t + i sin t, which she'll derive herself in Unit 10.

Look at what just happened. The derivatives of powers are powers: x³ → 3x². x² → 2x. x¹ → 1. x⁰ → 0. x⁻¹ → −x⁻².

Nothing in that list produces x⁻¹. Run it backwards: what has derivative 1/x? No power of x does — the power rule always lowers the exponent by one, and to land on −1 you'd need to start at 0, but x⁰ is constant with derivative 0.

There's a hole in the power rule, and ln x is what fills it. That's not a coincidence or a curiosity; it's why logarithms show up in integration constantly. Unit 6 will define ln x as an area under the hyperbola 1/x, which explains the coincidence properly.

Notice what happened: differentiating a transcendental function (arctan) produced a purely algebraic one (1/(1+x²)).

Run that backwards and you get the fact that ∫ dx/(1+x²) = arctan x — an innocent-looking rational function whose antiderivative is a trig function. There is no way to guess that from the integrand. It's a large part of why integration in Unit 6 is a bag of tricks rather than an algorithm, and it's why these two derivatives must be memorized in both directions.

The move is to treat y as an unknown function of x and differentiate. But if the curve isn't a function, what justifies pretending it is?

The answer is the Implicit Function Theorem: near almost any point on such a curve, a small enough piece of it is the graph of a function, even though the whole curve isn't. A circle isn't a function; the top half near (3,4) is. You're differentiating that local piece.

The exceptions are exactly the points where the curve is locally vertical. And this is the elegant part — the formula announces its own failure at precisely those points, because that's where its denominator hits zero. For the circle, dy/dx = −x/y blows up at y = 0, which is (±5, 0), the two points where the circle is vertical. The algebra knows where the theorem stops working.

BC never mentions any of this. It's worth having, because "why is this legal" is the obvious question and the textbook answer is silence.

Unit 5 asked what derivatives say about graphs. This unit asks what they say about the world. The calculus is mostly the same; what's being taught is modeling — turning a situation into an equation you can differentiate. That translation is the actual skill, and it's the one that transfers.

L'Hôpital's rule also lands here in most courses, because it needs derivatives. Full treatment below.

Speed is |v|, so speeding up means |v| is growing. If the object is moving left (v < 0) and acceleration is also negative, it's being pushed further left — moving faster in the negative direction. |v| grows. So negative acceleration can mean speeding up.

The clean statement: acceleration in the same direction as motion adds speed; acceleration opposing motion removes it. Same-sign, speeding up. Opposite-sign, slowing down. This is asked on nearly every AP exam and missed constantly.

In implicit differentiation, y is secretly a function of x. In related rates, everything is secretly a function of t. Same rule, different hidden variable. The curriculum presents them as separate topics; telling her they're one thing is worth doing.

Textbook related-rates problems are contrived because the honest ones need multivariable calculus. One that isn't: cumulative arrival and departure curves at a transit platform.

Riders arrive at rate λ(t); the cumulative count A(t) is its integral. Departures D(t) are a second curve. The vertical gap A − D is the queue length; the horizontal gap is an individual rider's wait; the area between the curves is total passenger-delay. The related rate — how fast is the queue growing? — is just A′(t) − D′(t).

Everything in that picture is a calculus object with a physical meaning she can point at, which the ladder is not.

You want a root of f. You have a guess x_n. Replace f by its tangent line there — the linear approximation again — and solve that for zero instead, since solving a line is easy.

The tangent is y = f(x_n) + f′(x_n)(x − x_n). Set y = 0 and solve for x: x = x_n − f(x_n)/f′(x_n). That's the formula, derived in one line. Then repeat from the new point.

Convergence is quadratic — the number of correct digits roughly doubles each step. That's why it's still the workhorse root-finder three and a half centuries later.

Near a, each function is well approximated by its tangent line. If f(a) = g(a) = 0, both tangent lines pass through zero there, so near a:

The ratio is f′(a)(x−a) / g′(a)(x−a), and the (x−a) cancels, leaving f′(a)/g′(a).

L'Hôpital is cancelling the common factor again — the same move as factoring in Unit 1, executed with tangent lines instead of algebra. That's exactly why it requires the 0/0 condition: without it the tangent lines don't both pass through zero and there's no shared factor to cancel.

(The honest proof uses the Cauchy Mean Value Theorem, a two-function version of MVT, rather than this hand-wave. The intuition is right and the machinery is Unit 5's.)

Next stop — multivariable calculus. With several inputs, "slope" is meaningless but "best linear approximation" still works: the derivative becomes a gradient vector pointing in the direction of steepest increase, and then a Jacobian matrix when the output is also multidimensional.

Then — optimization and machine learning. Gradient descent is literally "compute the gradient, step downhill, repeat" — Newton's method's cousin, run in ten million dimensions. Backpropagation is the chain rule applied through a composed function with millions of layers. If she is at all interested in AI, this is the honest answer to "what is calculus for": the entire training procedure of a neural network is Unit 3's chain rule plus Unit 4's linear approximation, at scale.

You can compute derivatives. Now: what do they tell you? This unit is the bridge from local information (the slope at a point) to global conclusions (the function is increasing on this whole interval).

That bridge is one theorem, and it's the one every student writes off as filler.

Students think MVT is a curiosity. It is the only thing licensing every inference from a derivative back to the function. Each of these claims is a corollary of MVT and is false without it:

• f′ > 0 on an interval ⟹ f is increasing there. Proof: take any two points p < q in the interval. MVT gives a c with f(q) − f(p) = f′(c)(q − p). Both factors are positive, so f(q) > f(p). Done.
• f′ = 0 everywhere ⟹ f is constant. Same argument: f(q) − f(p) = 0·(q−p) = 0 for every pair.
• Two functions with the same derivative differ by a constant. Apply the previous fact to their difference.

That last one is why the +C exists. In Unit 6 you'll write ∫f dx = F(x) + C and the C will be presented as a rule to remember. It isn't a rule — it's this theorem. And the fact that antiderivatives are unique up to a constant is precisely what makes the whole integral-as-antiderivative program coherent. Without MVT, the Fundamental Theorem doesn't work.

Fermat's Theorem: if f has a local extremum at an interior point c and f′(c) exists, then f′(c) = 0.

Why: suppose f′(c) > 0 at a local max. Then f is increasing through c, so points just to the right are higher — contradicting that c is a max. Same argument with the sign flipped rules out f′(c) < 0. The only survivor is 0.

The converse is false and this is the most common misunderstanding in the unit. f(x) = x³ has f′(0) = 0 but no extremum there — the curve flattens and keeps going. A critical point is a candidate, not a conclusion. You must always test.

f″ > 0 says f′ is increasing — the slope itself is getting larger. A curve whose slope keeps rising (from steeply negative, through zero, to steeply positive) is by definition curving upward. The second derivative isn't a new kind of object; it's the first derivative rule applied one level up.

This is also exactly why the second derivative test works: at a critical point the slope is zero, and if the slope is increasing through zero, it went from negative to positive — falling then rising — which is a minimum.

x⁴, −x⁴, and x³ all have f′(0) = f″(0) = 0, and they have a minimum, a maximum, and neither, respectively. The test simply can't distinguish them.

What the test is actually doing: checking whether the best quadratic approximation to f near c opens up or down. When f″(c) = 0 the quadratic is flat and carries no information — you'd have to look at the cubic term. That's Taylor series (Unit 10) leaking backwards into November.

The calculus in an optimization problem is trivial — differentiate a function and set it to zero, which she could do in week six. What's being tested is whether she can convert a paragraph of English into one function of one variable.

The structure is always the same: two quantities, one you're optimizing and one that's fixed. Find which is which. "Minimize material for a box holding 32 cubic units" — material is the objective, volume is the constraint. "Maximize area with 100 feet of fence" — area is the objective, perimeter is the constraint. Getting these backwards is the most common failure, and it happens before any calculus starts.

Next stop — the calculus of variations, which optimizes over a space of functions rather than numbers. The brachistochrone problem from Unit 9 is its founding question. The Euler–Lagrange equation is its central result.

Then — Lagrangian and Hamiltonian mechanics, which reformulate all of physics as "nature minimizes a certain integral." That reformulation is what makes quantum mechanics and field theory expressible at all. It's arguably the deepest idea reachable from a BC starting point.

The derivative half is complete. This is the second of the two operations — and it is introduced before the Fundamental Theorem on purpose, so she learns what an integral is before she learns the shortcut for computing one.

Good courses spend a painful week on Riemann sums for exactly this reason. Students hate it; it feels like arithmetic busywork when a shortcut is visibly coming. It's the correct pedagogical decision, and if her teacher rushes it, that's the gap to fill — because a student who thinks "integral = antiderivative" cannot make sense of anything in Units 8 or 9.

Leibniz's ∫ is an elongated S, for summa. The dx is the width of an infinitesimal slice. So ∫ f(x) dx reads literally: the sum of (height × width).

Once you see that, the notation stops being a symbol to memorize and becomes a description of the procedure. It also explains why the dx is not optional decoration — it's the width factor, and in Unit 6 substitution it will need to be converted like any other quantity.

Area is the picture, not the idea. The integral accumulates a product where one factor won't hold still.

• rate × time = distance — but only if the rate is constant. If it varies, integrate.
• force × distance = work — only if force is constant. If it varies, integrate.
• density × volume = mass. Varying density? Integrate.
• probability density × interval = probability.
• load × distance = passenger-miles; queue length × time = total delay.

Every one of those is "multiply two things where one of them varies." Teaching integration purely as area is why students later can't recognize an integral when it shows up as work or as expected value. The unit on the integrand is multiplied by the unit on dx — always check it.

A trapezoid connects the two endpoints with a straight chord. On a concave-up curve the chord lies above the curve, so the trapezoid includes extra area → overestimate.

The midpoint rectangle is subtler. Its top is a horizontal line at the midpoint height — but tilt that line to be tangent at the midpoint and the area doesn't change, because the triangle you add on one side exactly matches the one you remove on the other. And on a concave-up curve the tangent lies below the curve → underestimate.

That's a genuinely nice argument and it's the reason midpoint is more accurate than trapezoid (about twice as accurate), which surprises people who expect the fancier-looking method to win.

The average of n numbers is their sum divided by n. For a continuous function there are infinitely many values, so "sum" becomes an integral and "divide by how many" becomes divide by the length of the interval.

Rearranged: ∫_a^b f = f_avg·(b−a). That's a rectangle of height f_avg and width (b−a) with the same area as the region — the flat level the curve would need if you smoothed it out.

Unit 2 left a hole: nothing in the power rule produces 1/x, because the rule always lowers the exponent by one and you'd have to start at x⁰. Here's the resolution. Define:

Then everything about logarithms falls out of geometry:

• d/dx[ln x] = 1/x is immediate from the Fundamental Theorem (Unit 6) — no longer a mystery entry in a table.
• ln(ab) = ln a + ln b comes from the substitution t → at: the area from 1 to ab splits into the area from 1 to a plus the area from a to ab, and scaling maps that second piece exactly onto the area from 1 to b. The log law is a scaling symmetry of the hyperbola.
• e is simply the number where the accumulated area first reaches 1.

Napier invented logarithms in 1614 as a pure calculating aid — a way to turn multiplication into addition for astronomers. That they are also a fact about the area under y = 1/x was discovered by Grégoire de Saint-Vincent in the 1640s, and it genuinely stunned people. Two completely unrelated-looking things turned out to be one thing.

Everything so far has been construction. This is the payoff, and it is the centre of the subject. Two problems with no visible relationship — find the slope of a tangent and find the area under a curve — turn out to be inverse operations.

Nobody saw this for two thousand years. Archimedes computed the area under a parabola in 250 BC. Fermat and Descartes had tangent methods by the 1630s. Barrow, Newton's own teacher, had a geometric version of the connection around 1660 and did not recognize what he was holding. Newton and Leibniz did, independently, within about a decade of each other.

Say this to her explicitly. It's the most surprising thing she'll learn all year, and it is routinely taught as a computational rule in forty minutes.

Before: every area problem was a separate feat of genius. Archimedes needed one brilliant construction for the parabola, and it told you nothing about the hyperbola.

After: find an antiderivative, subtract at the endpoints. An entire class of genius-required problems collapsed into a procedure. That's the actual revolution — not new answers, but the mechanization of a problem type.

It's also why the subject is called calculus, which just means "a small pebble used for counting." The name advertises that it's a method of reckoning, not a body of results.

Unit 5's MVT proved: two functions with the same derivative differ by a constant. So once you have one antiderivative F, every other one is F + C, and there are no others hiding.

That's what makes FTC Part 1 well-defined. If you and I pick different antiderivatives, we differ by a constant — and when we subtract F(b) − F(a), the constant cancels. We get the same answer. Without MVT there'd be no guarantee of that, and the shortcut wouldn't be reliable.

FTC Part 2 guarantees every continuous function has an antiderivative: ∫_a^xf is one, and it exists whether or not you can write it down.

But most such antiderivatives cannot be expressed in elementary terms. e^−x² is the famous case. Liouville proved in the 1830s that no elementary antiderivative exists — not "nobody has found one," but there is none. Same for sin(x²), for (sin x)/x, for e^x/x.

And yet ∫_−∞^∞ e^−x²dx = √π, exactly. The bell curve — the most consequential function in all of statistics — cannot be integrated by any technique in her course, but its total area is a clean closed form. (The trick: square the integral, convert to polar coordinates. That's Calc III, but it's a fifteen-minute story.)

The deep asymmetry: differentiation is an algorithm — feed in any elementary function, turn the crank, get an elementary answer, always. Integration is not. It's pattern recognition and luck. That is why Unit 6 is a bag of tricks rather than a method, and it's worth telling her before she starts wondering what's wrong with her.

FTC reduced integration to antidifferentiation. But antidifferentiation has no algorithm — that's the asymmetry flagged at the end of Unit 6. So what follows is a bag of techniques, ordered roughly by how often each one works.

AB stops at u-substitution. Everything after it is BC-only, which is why the course accelerates here. Two further notes on scope: partial fractions is on the AP but only for non-repeating linear factors, and trigonometric substitution is not on the AP at all — many teachers cover it anyway.

Each technique is a differentiation rule run backwards. Recognizing which rule is being reversed is how you choose.

The chain rule says d/dx[F(g(x))] = F′(g(x))·g′(x). Read that equation right to left: anything of the form (function of g) times (derivative of g) is the derivative of a composite, so its antiderivative is that composite.

What to look for: a function and its own derivative both present in the integrand, up to a constant factor. The substitution just gives you a bookkeeping system for exploiting that.

It doesn't solve the integral. It trades one integral for another, and the whole art is choosing u so the trade is favourable — meaning ∫v du is easier than ∫u dv.

Rule of thumb: pick u to be the thing that gets simpler when differentiated. ln x becomes 1/x (much simpler). x becomes 1 (simpler). sin x becomes cos x (no simpler). e^x stays e^x (no simpler).

The standard mnemonic for the priority order of u is LIATE: Logarithmic, Inverse trig, Algebraic, Trigonometric, Exponential. Whichever appears earliest in that list, make it u. It's a heuristic, not a theorem, and it works most of the time.

Every polynomial with real coefficients factors into linear and irreducible-quadratic pieces. That's the Fundamental Theorem of Algebra (Gauss, 1799) doing hidden work — a theorem from a completely different branch of mathematics is what makes this integration technique reliable.

And every resulting piece integrates to a log or an arctan. So the whole class of rational functions is integrable in elementary terms, always. That's rare and worth noticing: it's one of the few places where integration does have an algorithm.

The boundary case p = 1 diverges. That razor-thin threshold — 1/x fails, 1/x^1.0001 succeeds — is the same threshold that will govern p-series convergence in Unit 10, and for the same reason. The integral test makes the connection explicit.

It's the growth hierarchy from Unit 1 again, deciding whether a tail shrinks fast enough.

Rotate y = 1/x for x ≥ 1 around the x-axis.

• Volume = π∫₁^∞x⁻²dx = π. Finite.
• Surface area ≥ 2π∫₁^∞x⁻¹dx = ∞. Infinite.

A solid you could fill with π cubic units of paint but could never finish painting. Torricelli found this in 1641 and it caused a real crisis — Hobbes thought it discredited the whole enterprise.

The resolution (mathematical surfaces have no thickness; real paint does) is less interesting than the mechanism: the two integrals sit on opposite sides of the p = 1 threshold. It's a physical illustration of a convergence boundary.

Next stop — vector calculus. Green's theorem, Stokes' theorem, and the Divergence theorem each say a version of "the integral over a boundary equals the integral of a derivative over the interior." They are all special cases of the generalized Stokes' theorem, ∫_∂Ωω = ∫_Ωdω.

Then — differential forms and differential geometry, where that statement is the definition rather than a theorem, and then general relativity, which is differential geometry with physics attached.

The curriculum places this here because you need integration techniques first. But historically, this is what calculus was invented for.

Newton didn't build the subject to find areas. He built it because F = ma is a differential equation, and he needed to solve it to derive Kepler's elliptical orbits from an inverse-square law of gravitation. Every fundamental law of physics is a differential equation. Maxwell's equations, the heat equation, Schrödinger's equation, the Navier–Stokes equations — all of them.

Worth telling her explicitly: everything for the last six months was scaffolding for this.

A differential equation is a local rule: wherever you are, here is your slope. Solving it means finding the global trajectory consistent with that rule at every point.

Local law → global behavior. That's the same theme as the Mean Value Theorem in Unit 5, and it's arguably the central move of the entire subject.

They show that solutions exist and have structure even when you can't write them down — which is the honest situation for the overwhelming majority of differential equations. The handful she'll learn to solve algebraically are the rare, tame cases.

Reading a slope field is also its own AP skill: identify equilibrium solutions (rows of horizontal ticks, where dy/dx = 0), spot where solutions increase or decrease, and match a field to an equation by testing a couple of points.

The move treats dy/dx as a fraction and multiplies both sides by dx. In a limit-based framework, dy/dx is not a fraction — it's a single symbol denoting a limit — so this looks like nonsense.

The proper justification is the chain rule: if H′(y) = 1/h(y), then d/dx[H(y)] = H′(y)·dy/dx, and the manipulation is really an application of the chain rule in reverse. Same relationship u-substitution has to the chain rule.

Her instinct is right and her tools are the problem, and that's the interesting part of the story. Bishop Berkeley attacked exactly this in 1734 — he called infinitesimals "the ghosts of departed quantities" and pointed out that dx has to be both zero and not-zero for the arguments to work. He was correct, and nobody could answer him for 130 years. Weierstrass's limits (Unit 1) dodged the problem rather than solving it. Then Abraham Robinson built nonstandard analysis in 1966 and showed infinitesimals were legitimate objects all along — Leibniz had simply been three centuries ahead of the logic needed to justify him.

Read the equation in words: the rate of growth is proportional to the current amount. That's what "exponential" means. It's not a statement about a formula — it's a statement about a mechanism, and the formula is what the mechanism forces.

This is also Unit 2's fact that e^x is its own derivative, restated. Interest compounds because the interest earns interest; populations grow because organisms produce more organisms; radioactive decay is proportional to how much is left because each atom decays independently.

Exponential growth is unphysical for a population — nothing grows without bound forever. The logistic model multiplies by an extra factor that measures how much room is left.

• When P is small, (1 − P/M) ≈ 1 and growth is essentially exponential.
• As P approaches M, the factor approaches 0 and growth stalls.

M is the carrying capacity. It's also an equilibrium solution: if P = M exactly, dP/dt = 0 and the population sits there forever. P = 0 is the other equilibrium — unstable, since any small population grows away from it.

Verhulst introduced this in 1838, explicitly to answer Malthus's prediction of unbounded population growth.

Stand at a known point. The equation tells you the slope there. Follow the tangent line for a short step h and land somewhere new. Recompute the slope. Repeat.

This is linear approximation applied over and over — the same idea as Newton's method, redeployed. Also the same idea as a Riemann sum: approximate a curved thing by many small straight pieces.

Error is O(h): halving the step size only halves the error, which is poor. Real solvers use Runge–Kutta, where halving the step cuts the error by sixteen.

The Lorenz system (1963) is three simple, fully deterministic differential equations modelling atmospheric convection. Their solutions never repeat and depend so violently on the starting values that prediction fails within days.

Lorenz found it by accident: re-running a simulation, he typed 0.506 instead of the stored 0.506127 and got a completely different forecast. Chaos falls out of equations no more complicated-looking than the ones she's solving.

And a hard limit worth knowing: the three-body problem has no closed-form solution. Two bodies orbiting under gravity, Newton solved exactly. Add a third and no formula exists — Poincaré proved it in 1889, in a competition held for a Swedish king's birthday. Some equations simply cannot be "solved," and that discovery founded the modern study of dynamical systems.

Next stop — ODEs proper: systems, linear equations with integrating factors, second-order equations, and phase portraits, which show all solutions at once as flows in a plane. This is where linear algebra becomes unavoidable — solving a linear system of DEs is an eigenvalue problem.

Then, two forks. Dynamical systems and chaos — stability, bifurcations, strange attractors, the Lorenz system done properly. Or partial differential equations — the heat equation, wave equation, Laplace's equation, which is most of mathematical physics and requires Fourier series, which is the natural sequel to Unit 10.

Next stop — numerical analysis. Everything she computes by hand, done by machine with rigorous error bounds. Euler → Runge–Kutta. Newton → the whole of nonlinear optimization. Riemann sums → Gaussian quadrature. Also the study of when these fail: stiffness, instability, catastrophic cancellation.

Practical note: this is the fastest route from "calculus is beautiful" to "calculus does things," and it's approachable right now. A hundred lines of Python implementing Euler's method on a system she chooses, and watching it diverge from the true solution as the step size grows, teaches more about differential equations than a chapter of exercises.

Techniques are in hand; deploy them. Conceptually this unit adds almost nothing new — the integrals are usually easy. What's being taught is a method of setup, and that method is the entire content.

Use dy when the region is more naturally described by "right curve minus left curve" — typically when a single vertical slice would change which curve bounds it partway up, forcing you to split. One dy integral often replaces two dx integrals.

The cost is that you must solve the equations for x in terms of y, which isn't always possible. Try both mentally before committing.

Take a tiny piece of curve. It's essentially straight, so it's the hypotenuse of a right triangle with legs dx and dy:

Factoring dx out of the root is the only step. Then sum the ds pieces.

Worth knowing: ds² = dx² + dy² is the metric, and generalizing it to curved spaces is literally how differential geometry — and then general relativity — begins. The arc length formula is a first glimpse of a very deep object.

Next stop — continuous probability. A density is a function; probability is its integral; expected value is ∫x·p(x)dx; variance is another integral. The bell curve is e^−x²/2 normalized, and its non-elementary antiderivative is exactly why every statistics text has a Z-table.

Then — stochastic calculus, which does calculus on paths that are continuous everywhere and differentiable nowhere. Those paths are Weierstrass's function from Unit 1, made physical. Itô's lemma is the chain rule corrected for the fact that the path has no derivative, and the Black–Scholes equation is a PDE built on it. Unit 1's "lamentable plague" turns out to be the foundation of modern finance.

This is the first genuine widening of what a "function" is, and it exists to break an assumption that has been quietly limiting everything since Unit 1: that curves are graphs of y = f(x).

They aren't. Circles aren't. Orbits aren't. Anything that loops, crosses itself, stops, or reverses isn't. The machinery all still works — you just have to describe the curve differently.

It's placed here because it needs both differentiation and integration in hand, and because Unit 10 will also require the idea that functions can be built in unfamiliar ways.

It's how Newton thought — kinematically, in terms of quantities flowing in time. And it's strictly more general: every function y = f(x) can be parametrized as x = t, y = f(t), but most parametric curves cannot be written as functions.

Crucially, the curve and the parametrization are different objects. The unit circle is (cos t, sin t), but it's also (cos 2t, sin 2t) — same path, traversed twice as fast. Speed, direction, and starting point are properties of the parametrization, not of the curve.

It's literally ∫ |velocity| dt — distance equals speed integrated over time, the most ordinary fact in physics.

Compare Unit 8's Cartesian version, ∫√(1 + (dy/dx)²)dx. Same Pythagorean origin (ds² = dx² + dy²), but the parametric form is symmetric in x and y and doesn't break when the curve goes vertical. The parametric formula is the real one; the Cartesian one is the special case x = t.

Some situations have rotational symmetry, and forcing them into a rectangular grid produces ugly algebra. A circle of radius 3 is x² + y² = 9 in Cartesian and simply r = 3 in polar.

The best example: Kepler's second law says a planet sweeps equal areas in equal times. In Cartesian coordinates that's a mess. In polar it's ½r²(dθ/dt) = constant — which is exactly the polar area formula differentiated, and it's conservation of angular momentum. The physics becomes visible when the coordinates match the geometry.

It's the on-ramp to multivariable calculus. Once position, velocity, and acceleration are vectors, everything in mechanics can be written in one line instead of two, and the same notation survives into three dimensions and beyond.

It also makes a distinction visible that scalars hide: acceleration can change a particle's direction without changing its speed. Uniform circular motion has constant speed and constant-magnitude acceleration pointing always toward the centre. In one dimension that's impossible; in two it's the most common motion in the universe.

A bead slides under gravity from A down to a lower point B. Which curve gets it there fastest?

Not the straight line. The answer is an upside-down cycloid — the path traced by a point on the rim of a rolling wheel, which is naturally parametric: x = t − sin t, y = 1 − cos t.

Johann Bernoulli posed it as a public challenge in 1696. Newton, then 55 and running the Royal Mint, received it in the evening, solved it overnight, and published anonymously. Bernoulli recognized the author immediately: "tanquam ex ungue leonem" — one knows the lion by its claw.

The same cycloid is also the tautochrone: a bead released from any point on it reaches the bottom in the same time. Huygens used that to design a pendulum clock in 1659, before the calculus existed to explain why it worked.

Solving it properly requires the calculus of variations — optimizing over a space of functions rather than over numbers. That framework is the whole of Lagrangian mechanics, and it's what a sophomore who likes this material might chase next.

It comes last because it needs everything: limits (of sequences now), integration (the integral test), improper integrals (the p-threshold), and derivatives of every order (Taylor).

Warn her about the three weeks. The convergence-test section feels like an unrelated course — a zoo of arbitrary rules about infinite sums with no visible purpose. Then Taylor series arrives and retroactively justifies all of it: the tests were the tools needed to know when an infinite polynomial is legitimate. Knowing the payoff is coming makes the zoo tolerable.

It's also the hardest unit in BC and the largest slice of the exam. Budget accordingly.

An infinite sum is not a sum. You cannot perform infinitely many additions. What you can do is perform each finite prefix and ask where those results are heading.

Zeno's paradoxes are precisely this confusion. To cross a room you must first cross half, then half the remainder, and so on — infinitely many steps, so (Zeno concluded) motion is impossible. The resolution is that ½ + ¼ + ⅛ + ⋯ has partial sums ½, ¾, ⅞, … converging to 1. Infinitely many terms, finite total. That took 2,400 years to state properly, and the reason is that nobody had the concept of a limit.

The tests look like an arbitrary collection. They're all asking one thing: does the tail shrink fast enough? Each test is a different way of measuring "fast enough," and each is really the Unit 1 growth hierarchy in disguise.

A conditionally convergent series can be rearranged to sum to any real number you choose. Any number at all. Or to +∞, or to −∞.

The mechanism: the positive terms alone diverge and the negative terms alone diverge. So take positive terms until you exceed your target, then negatives until you fall below, then positives again — you can steer the partial sums anywhere.

For conditionally convergent series, addition stops being commutative. That's a genuinely disturbing fact about infinity, and it's the entire reason the absolute/conditional distinction exists. Absolutely convergent series are immune — you can reorder them freely.

Look at the first few terms:

• Degree 0: f(a) — a constant approximation.
• Degree 1: f(a) + f′(a)(x−a) — that's linear approximation from Unit 4, verbatim.
• Degree 2: adds f″(a)(x−a)²/2 — the quadratic whose opening direction the second derivative test checks.
• Keep going and you get the whole function.

Every approximation idea in the course was a truncated Taylor series. Linear approximation, differentials, the second derivative test, Newton's method, Euler's method, L'Hôpital — all of them are this, cut off early. That's why the Unit 2 framing "the derivative is the multiplier in the best linear approximation" was worth carrying since September: this unit is where it pays.

Sine is odd, so only odd powers survive; cosine is even, so only even ones. Their coefficients alternate because their derivatives cycle with period 4 (Unit 3 flagged this). e^x has no alternation because every derivative is itself.

Everything else in the course is obtained from these by substitution, differentiation, integration, or multiplication — and doing it that way is far faster and far less error-prone than computing derivatives one at a time.

Students skip the error bound because it's fiddly. It is the part that makes the whole enterprise useful rather than decorative — without it, a truncated series is a guess, and with it, it's a guarantee.

Read the formula: it's the next term you didn't include, with the derivative replaced by its worst-case value on the interval. Nothing more.

The alternating series bound is even simpler and worth preferring when it applies: for an alternating series with decreasing terms, the error is smaller than the first omitted term. One term, no maximization.

This is the largest unit on the exam and reliably the hardest. What follows is the additional depth that the earlier pass compressed: a proper decision procedure, the root test, the tests she'll misapply, and the Taylor manipulations that come up every year.

What is 1 + ¼ + ⅑ + 1/16 + ⋯ ? Posed in 1650, it resisted the Bernoullis for decades — they could prove it converged and could not find the value.

Euler solved it in 1735 at age 28 and got π²/6. Where does π come from in a sum of reciprocal squares? His method — factoring sin x as an infinite product from its roots, treating a transcendental function as a giant polynomial — was completely unjustified by the standards of the time and completely correct. It made him famous across Europe overnight.

1 − ½ + ⅓ − ¼ + ⋯ = ln 2. The same terms, without the signs, diverge to infinity. And by Riemann's theorem, those same terms rearranged can be made to sum to 7, or to π, or to anything you name.

Write out the Maclaurin series for e^x and substitute ix. The powers of i cycle 1, i, −1, −i, with period 4 — the same period as the derivative cycle of sine and cosine, which is not a coincidence.

Separate the real and imaginary terms. The real ones are exactly cosine's series. The imaginary ones are exactly sine's:

Set x = π, and since cos π = −1 and sin π = 0:

Five fundamental constants, three operations, no slack. And it isn't mysticism — it's three series sitting next to each other, and she can do the derivation herself in about six lines.

Save this for the last week. It's the best possible ending to the year, and it only works as an ending if the three series are already familiar.

Next stop — Fourier series. Taylor series builds functions out of powers; Fourier builds them out of sines and cosines. It turns out to be the better basis for anything periodic, and it is the mathematical foundation of signal processing, audio and image compression, MRI reconstruction, and the JPEG format. The conceptual leap — that a function can be decomposed into frequencies — is one of the most consequential ideas in applied mathematics.

Then — complex analysis. Taylor series over the complex numbers behave far better than over the reals: a complex-differentiable function is automatically infinitely differentiable and automatically equal to its Taylor series. Facts that are fussy in real calculus become clean. It also explains something Unit 10 leaves mysterious: why 1/(1+x²) has radius of convergence 1 despite the function being perfectly well-behaved everywhere on the real line. The answer is that it blows up at x = ±i, and the radius of convergence is the distance to the nearest complex singularity. That's a genuinely satisfying loose end to be able to tie off.

BC is a specific set of choices about what to include, and those choices aren't universal. Knowing what other serious programs do tells you where BC is thin, what a college course will assume she has, and which gaps are worth filling if she stays interested.

Curricula change; verify anything decision-relevant against current official documents.

None of these are hard once she has BC. Trig substitution is two weeks; hyperbolic functions are an afternoon; integrating factors are a single technique. They're omissions of coverage, not of capability.

The practical risk is placement. If she takes BC credit and jumps into Calc III or a linear-algebra-and-differential-equations sequence, she may hit trig substitution or integrating factors assumed as background. Worth a summer afternoon each, not a course.

The genuine gap is proof. BC assesses computation and justification-in-a-sentence, not proof. A student who goes on to real analysis meets a different subject.

She spent two years in Russian-method algebra, and the tradition has a distinct philosophy that will shape how BC feels to her.

The Russian approach front-loads algebraic fluency and problem-solving over technique coverage. Students meet fewer named methods and more hard problems requiring the methods they have. Proof and derivation appear early and routinely. Limits are typically treated more rigorously and earlier. There's an explicit culture of problems that cannot be solved by pattern-matching.

Predicted consequences for her:

• The algebra in BC will never be the bottleneck. She'll be faster than her classmates at the manipulation and may find drill sections tedious.
• She may find AP free response strangely easy — the problems are structured and scaffolded compared with olympiad-style work.
• She may find the justification requirements irritating rather than difficult: being asked to write "f′ changes from positive to negative, therefore…" can read as insultingly obvious to a student trained to prove things. It's worth telling her that this is an exam convention, not a claim about what's hard, and that the points are real regardless.
• The place she's most likely to be genuinely challenged is series, because it rewards a kind of pattern-recognition-under-constraint that's different from algebraic power.