Powers, logarithms, trigonometry, functions, the numerical-analysis toolkit — and now calculus, probability and statistics too. No fluff, no proofs for their own sake: intuition first, then a worked example you can follow line by line. The "oh right, that's what it means" course.
Everything later leans on powers and their rules. A power is just repeated multiplication — and once you see the handful of rules, they never need memorising again because they all come from that one fact.
an means "multiply a by itself n times." So 23 = 2×2×2 = 8. The a is the base, the n is the exponent (or power, or index).
Why? 22·23 is (2×2)×(2×2×2) — five 2's — so you add the exponents. Dividing cancels multiplications, so you subtract. A power of a power multiplies. Three special cases fall straight out:
Write any number as (a number between 1 and 10) × 10power. The power counts how many places the decimal point moves.
Brackets, Orders (powers/roots), Division & Multiplication, Addition & Subtraction. (Americans say PEMDAS — same thing.) Multiplication before addition is the one that trips everyone:
A logarithm sounds intimidating and is genuinely simple: it's the question "what power?" If powers ask "what is 10 cubed," logs ask the reverse — "10 to what gives 1000?"
So log10(1000) = 3 because 103 = 1000. And log2(8) = 3 because 23 = 8. That's the whole idea — a log is an exponent in disguise. Two facts follow immediately: logb(1) = 0 always (any base to the power 0 is 1), and logb(b) = 1.
Two bases dominate. Base 10 (log) is the everyday one. Base e ≈ 2.718 — a special constant that shows up wherever things grow continuously — gives the natural log, written ln. So ln(e) = 1 and ln(e3) = 3. When a formula uses ln or log without a base, it almost always means natural log.
These mirror the power rules exactly (because logs are exponents). The first is the historically famous one: before calculators, logs let people multiply huge numbers by adding — that's what slide rules and log tables did.
Trigonometry connects angles to lengths. It starts with right triangles and three ratios, then the unit circle turns those ratios into something that works for any angle and gives you waves for free.
In a right triangle, label the sides relative to an angle θ: the hypotenuse (longest, opposite the right angle), the opposite (facing θ), and the adjacent (touching θ). Then:
Two ways to measure an angle. Degrees are familiar (a full turn is 360°). Radians are the mathematician's unit, based on the circle itself: a full turn is 2π radians. The bridge you need:
Draw a circle of radius 1. A point on it at angle θ has coordinates (cos θ, sin θ) — that's the definition that frees trig from triangles and lets angles go past 90°, negative, all the way around. tan θ is the slope of the radius line. Drag the point:
| θ | 0° | 30° | 45° | 60° | 90° |
|---|---|---|---|---|---|
| sin | 0 | 0.5 | 0.707 | 0.866 | 1 |
| cos | 1 | 0.866 | 0.707 | 0.5 | 0 |
| tan | 0 | 0.577 | 1 | 1.732 | ∞ |
As the point goes round and round, its height (sin) rises and falls, rises and falls — tracing the smooth wave you've seen everywhere. That repetition (every 360°/2π) is why sin and cos describe anything cyclical: sound, tides, seasons, alternating current.
A function is just a rule that turns an input into exactly one output — a machine with a slot and a chute. Graphs are the picture of that rule. Getting fluent at reading the three common shapes makes everything else legible.
Write f(x) for "the output when the input is x." The rule f(x) = 2x + 1 takes 3 to 7. The only requirement to be a function: one input gives one output — no ambiguity.
The slope m is "rise over run" — how much y changes for each step in x. Slope 2 means "up 2 for every 1 across." Constant slope is what makes it straight.
y = a·x2 + b·x + c makes a U-shaped curve (a parabola). The x2 term is what bends it; if a is negative the U flips upside down. This is the shape of a thrown ball's path.
The crucial contrast with linear. Linear adds the same amount each step; exponential multiplies by the same factor each step.
Given a graph, ask one question: does it go up by a fixed amount (straight line, linear), curve gently (quadratic), or bend sharply upward and keep steepening (exponential)? Naming the shape tells you which rule is behind it — and, from Day 2, its log turns that exponential back into a readable straight line.
Real calculation almost never gives clean answers — it gives numbers you must round, and every rounding introduces error. Numerical analysis is the discipline of knowing how much error, and keeping it under control. This is where "the math" meets the real world.
Two different rulers. Decimal places (dp) counts digits after the point. Significant figures (sf) counts meaningful digits from the first non-zero one — better for expressing precision, because it doesn't care where the decimal point sits.
Absolute error is the raw gap. Relative error is that gap as a fraction of the thing itself — usually the one you care about, because being off by 1 is trivial on a million and catastrophic on a two.
Try it — pick a value and a rounding, and watch both errors respond:
Round early and often, and small errors pile up. Add a thousand numbers each rounded to 2 dp and your total can be off by far more than 0.005. The practitioner's rule: keep full precision through the calculation, round only the final answer. Rounding intermediate steps is one of the most common sources of wrong results in real spreadsheets and code.
Most equations that matter can't be solved with a neat formula — there's no tidy expression for many roots, integrals, or the value of √2. So instead of solving exactly, we approximate, and often we iterate: guess, improve, repeat until the answer stops changing. This is how calculators and computers actually compute.
Zoom in on any smooth curve and it looks like a straight line. So near a point you know, you can estimate nearby values by following the tangent's slope:
When one estimate isn't enough, feed the answer back in and improve it. The classic is the Babylonian method for square roots, known for 3,000 years: to find √N, repeat
The intuition: if your guess is too small, N/x is too big, and vice versa — so their average is closer than either. Watch it converge on √N in a handful of steps:
You trade an exact answer you can't get for an approximate answer you can, plus a handle on the error. Every method balances three things: how accurate (how small the error), how fast it converges (how many steps), and whether it converges at all (some iterations run away to infinity). Knowing those three questions is most of the subject.
That's the numerical craft: rounding, error, and iteration. Next we add the three big pillars of higher math — change, chance, and data.
Calculus has a fearsome reputation and two friendly ideas at its heart. The derivative measures how fast something is changing right now. The integral adds a changing quantity up into a total. Almost everything else is technique on top of those two.
Think of a car's speedometer. Average speed is easy — distance ÷ time over a stretch. But the speedometer shows your speed at this instant. The derivative is exactly that: take the average rate of change over an interval, then shrink the interval to nothing.
Graphically: pick two points on a curve and draw the line through them (a "secant") — its slope is the average rate. Slide the points together and the secant becomes the tangent, whose slope is the instantaneous rate. That's the derivative, written f′(x) or dy/dx.
"Bring the power down in front, then knock one off the power." So x² → 2x (which is 6 at x=3, matching above), x³ → 3x², and any flat constant has slope 0. Drag the point along the curve and watch its tangent slope:
The integral runs the other way: it adds up a changing quantity, and geometrically that's the area under the curve. If a graph shows speed over time, the area beneath it is the distance travelled.
Differentiate a total and you get the rate; integrate a rate and you get back the total. Speed is the derivative of distance; distance is the integral of speed. That mirror-image relationship is the Fundamental Theorem of Calculus — and it's why the two ideas are always taught together.
Probability assigns a number between 0 and 1 to how likely something is: 0 means impossible, 1 means certain, 0.5 means even odds. A handful of rules let you combine those numbers correctly — which is harder than it looks, because intuition about chance is famously unreliable.
When outcomes are equally likely, probability is just counting:
One die: P(roll a 4) = 1/6. A deck: P(draw a heart) = 13/52 = 1/4.
Often far easier to compute the opposite. P(not a six) = 1 − 1/6 = 5/6. "At least one" problems are almost always easier as 1 − P(none).
Conditional — P(A | B), "the probability of A given B is known" — because information changes the odds. Given a die came up even (2, 4, 6), the chance it's a 2 is now 1/3, not 1/6: knowing "even" shrank the world of possibilities.
Expected value — the long-run average outcome, each value weighted by its probability:
The bridge between "the probability is 0.5" and reality: as you repeat a trial more and more, the observed fraction closes in on the true probability. A few flips can be lopsided; ten thousand won't be. Flip a coin and watch it settle:
Where probability reasons forward from known odds to outcomes, statistics reasons backward — from a pile of observed numbers to what they tell you. Three questions cover most of it: where's the centre, how spread out is it, and what's the shape?
The median is robust to outliers; the mean is not. This is exactly why incomes are reported as a median — a handful of very high earners would make the mean wildly unrepresentative of a typical person. Choosing the right measure of centre is half of using statistics honestly.
Two datasets can share a mean and be nothing alike — one tightly clustered, one wildly scattered. Standard deviation captures that: the typical distance of a value from the mean.
Astonishingly many quantities — heights, measurement errors, exam scores — pile up symmetrically around the mean in a bell curve. When they do, the standard deviation tells you almost everything, via one rule:
Powers and roots, logs as their inverse, trig from triangles to waves, functions as rules with shapes, the numerical craft of rounding, error and iteration — and now calculus for change, probability for chance, and statistics for data. That's the working mathematics most technical reading assumes, and the exact foundation the Transformer-math course was quietly standing on.