Road to A*

Mathematics, and what it is actually doing.

An A* is not the syllabus learned harder. It is being able to see the structure of a problem before you touch it, knowing why the methods work, and getting somewhere when nobody has told you which one to use.

Getting better is not about knowing more.

It is about becoming harder to surprise. Both subjects do that, and they do it differently enough that having both is worth more than being good at either.

MathematicsYou are here

UnderstandStop accepting a method as a rule. Ask what makes it true, and the exceptions stop being things to memorise.
SolveMeet problems that do not announce their method, and build the habit of looking for structure before calculating.
ExtendGo past the syllabus into proof, and find out what a mathematical argument has to do to count as finished.

Economics

ExplainReplace assertion with mechanism — a chain of steps each of which somebody could argue with.
EvaluateCompare two effects and decide which is larger, and say what that turns on. Not a paragraph beginning "however".
JudgeReach a conclusion that chooses, with the condition under which it would flip attached to it.

Road to A* Economics

What actually carries across

They ask the same first question

Both reward “why does this hold?” over “what is the rule?” A student who asks it in one subject usually starts asking it in the other.

Both make you choose

A hard maths problem does not say which method; a hard economics question does not say which model. Choosing, and being able to defend the choice, is the same skill twice.

They disagree about certainty

Maths ends in proof: the argument is finished or it is not. Economics almost never gets that, so it substitutes conditions — this holds while capacity is spare, while demand is inelastic. Learning where each standard applies is more useful than either alone.

That is what “transferable reasoning” actually means here — three specific habits, not a general claim about critical thinking.

Where are you?

Not levels — routes. Most people move between them depending on the topic, and being fluent in calculus while stuck on proof is completely normal.

I have just finished GCSEs

The gap between GCSE and A-level is mostly algebraic fluency. Find out which parts of GCSE stop being routine and start being load-bearing.

What actually carries over Why Maths opens the most doors

I am starting A-level Maths

Get the shape of the whole subject before you meet it topic by topic. Knowing what connects to what makes the second year far easier than the first.

The subject in ten families How the topics join up

I know the content but hard questions defeat me

This is the most common place to be stuck, and it is not a content problem. It is that unfamiliar questions do not announce which method they want.

Stop asking which formula What your mistakes are telling you

I want to go past the syllabus

Not harder arithmetic — different mathematics. Proof, structure, and the ideas A-level gestures at without developing.

Why the methods work Where it goes next

I am aiming at TMUA or olympiad problems

Problems where the difficulty is seeing what is going on, not executing a method. Almost none of it is content beyond A-level.

The approaches worth having What the TMUA actually is

What carries over from GCSE

Not a recap. These are the parts of GCSE that stop being topics and become the handwriting — the things you will use silently in every question for two years.

Algebraic manipulation

Rearranging, factorising and simplifying without thinking about it. At GCSE algebra is the question. At A-level it is the handwriting — every calculus, trigonometry and mechanics question is three lines of new content wrapped around ten lines of algebra you are expected to do silently.

Where it goes wrong. Treating (a + b)² as a² + b², and its relatives: √(a + b) ≠ √a + √b. Squaring and rooting do not distribute over addition. Almost every version of this error is the same error.

If a page of algebra takes you longer than the thinking that produced it, that is the thing to fix first.

Indices and surds

The index laws, negative and fractional indices, and rationalising. Differentiation of xⁿ only becomes useful when you can write 1/√x as x^−1/2 without pausing. A surprising number of A-level calculus errors are index errors.

Where it goes wrong. Reading x⁻¹ as "negative" rather than "reciprocal". A negative index is a direction on the multiplication scale, not a sign on the answer.

The index laws are the whole reason logarithms exist — see why logs turn multiplication into addition.

Functions and graphs

What a function does, domain and range, and how transformations move a graph. A-level asks you to read a situation off a shape. If you can picture y = f(x − 3) + 2 without plotting points, half of trigonometry and all of curve sketching become cheap.

Where it goes wrong. Getting the direction of horizontal shifts backwards. f(x − 3) moves the graph right, because you now need a larger x to feed the function the same input.

Transformations are the first place mathematics rewards seeing over calculating.

Quadratics

Factorising, the formula, completing the square, and the discriminant. Quadratics are the model for every later structure: a general form, a canonical form, and a criterion that tells you the nature of the solutions before you find them.

Where it goes wrong. Learning completing the square as a separate technique for a particular question type, rather than as the thing that generates the formula and the vertex and the discriminant all at once.

See why completing the square is the whole of quadratics.

Trigonometric relationships

The exact values, the identities, and what sine and cosine actually measure. Trigonometry stops being about triangles and becomes about periodic behaviour. If sine is still "opposite over hypotenuse" to you and not "height on the unit circle", the graphs will feel arbitrary.

Where it goes wrong. Losing solutions when solving trigonometric equations, because sin⁻¹ on a calculator returns one angle and the equation has infinitely many.

The unit-circle definition is what makes sin²θ + cos²θ = 1 obvious — it is Pythagoras on a radius of 1.

Proportional reasoning

Direct, inverse and combined proportion, and reading rates correctly. Every rate of change, every mechanics question and every growth model is proportional reasoning with more notation. Students who are comfortable here find calculus much less strange.

Where it goes wrong. Confusing "grows by a fixed amount" with "grows by a fixed factor" — the difference between an arithmetic and a geometric sequence, and between linear and exponential growth.

The compound-growth tool on Tools is this distinction made visible.

The subject in eleven families

Exam boards order these differently and split the applied content in different ways, but the mathematics is common to all of them. For each: what you have to be able to do, the idea underneath it, and the misconception that costs most marks.

Proof and mathematical argument

A proof is not a longer answer. It is an argument that leaves no case unexamined. This is the strand examiners use to separate students, and the one most revision material skips.

You need to be able to: Direct proof, proof by exhaustion, disproof by counterexample, and the language of "if", "only if" and "therefore".
The misconception: Verifying a claim for three values and calling it proved. Three cases are evidence; they are not an argument.
Where it connects: Everything. Proof is the thread that runs through all the other families.

Algebra and functions

A function is a rule with a domain. Most "impossible" A-level questions become routine once you ask what the domain is doing.

You need to be able to: Polynomials, the factor and remainder theorems, partial fractions, inequalities, modulus, composite and inverse functions.
The misconception: Multiplying an inequality by something that might be negative, which silently flips the direction.
Where it connects: Feeds directly into calculus — you cannot differentiate what you cannot rearrange.

Coordinate geometry

Geometry expressed as algebra. The circle theorems you learned at GCSE reappear as conditions on equations.

You need to be able to: Straight lines, circles, tangents and normals, parametric equations.
The misconception: Forgetting that a tangent to a circle is perpendicular to the radius at the point of contact — which turns most circle questions into one line.
Where it connects: Meets calculus at tangents and normals; meets vectors at lines in space.

Sequences and series

Sequences are the discrete version of functions, and the binomial expansion is a counting fact rather than an algebraic accident.

You need to be able to: Arithmetic and geometric sequences, sigma notation, binomial expansion, recurrence relations.
The misconception: Using the sum to infinity of a geometric series without checking |r| < 1. The formula is meaningless otherwise.
Where it connects: Geometric series are compound growth; see how the same maths behaves over a working life.

Trigonometry

Sine and cosine are coordinates on the unit circle. Every identity you are asked to memorise is a geometric fact about that circle written algebraically.

You need to be able to: Radians, exact values, identities, the sine and cosine rules, compound and double angle formulae, small-angle approximations.
The misconception: Working in degrees when the question is in radians — which quietly breaks every calculus result, because d/dx(sin x) = cos x is only true in radians.
Where it connects: Radians exist precisely so that calculus works cleanly.

Exponentials and logarithms

A logarithm is an index. e is the base at which a quantity grows at exactly its own size.

You need to be able to: Laws of logs, solving exponential equations, e and natural logarithms, exponential models.
The misconception: Applying log(a + b) as though it splits. It does not. Only products, quotients and powers behave.
Where it connects: The reason e^x is its own derivative, and why it appears in every growth model.

Differentiation

A derivative is a limit of gradients of chords. Everything else is bookkeeping.

You need to be able to: First principles, standard derivatives, chain, product and quotient rules, stationary points, connected rates of change, implicit differentiation.
The misconception: Finding a stationary point and asserting it is a maximum without testing. The second derivative being zero tells you nothing on its own.
Where it connects: Optimisation is where this meets economics — marginal cost is a derivative.

Integration

Integration accumulates. The link between accumulation and antidifferentiation is the deepest result at A-level and is usually stated rather than explained.

You need to be able to: Reverse of differentiation, definite integrals, areas, substitution, by parts, differential equations.
The misconception: Losing the constant of integration, and losing the fact that area below the axis contributes negatively to a definite integral.
Where it connects: See why an area is found by antidifferentiating.

Vectors

Vectors let you do geometry without coordinates and coordinates without pictures. The scalar product converts a geometric question about angle into an arithmetic one.

You need to be able to: Vectors in two and three dimensions, magnitude, unit vectors, the scalar product, geometric problems.
The misconception: Treating the position vector of a point and the displacement between two points as the same object.
Where it connects: Mechanics is vectors with units attached; university linear algebra starts here.

Probability and statistics

Statistics is inference under uncertainty. A hypothesis test does not tell you whether something is true; it tells you how surprised you should be.

You need to be able to: Sampling, conditional probability, the binomial and normal distributions, correlation and regression, hypothesis testing.
The misconception: Reading a significance level as the probability that the hypothesis is false. It is the probability of data this extreme if the hypothesis were true — which is a different statement.
Where it connects: The same reasoning underlies how risk is priced.

Mechanics

Mechanics is calculus with a physical interpretation and modelling assumptions you are expected to notice and criticise.

You need to be able to: Kinematics, forces, Newton’s laws, moments, projectiles, variable acceleration.
The misconception: Using the constant-acceleration equations when acceleration is not constant. If the question gives acceleration as a function of time, they want calculus.
Where it connects: The clearest place to see modelling assumptions doing real work.

Why the methods work

A-level teaches these as rules to apply. Every one of them has a reason, and the reasons are within reach — most take a paragraph. Where a full proof is genuinely beyond A-level, this says so rather than faking one.

01Why does differentiation give a gradient?Because it is the gradient of a chord, measured as the two points are brought together.

Take any two points on a curve, (x, f(x)) and (x + h, f(x + h)). The gradient of the straight line joining them is exact, not approximate:

[f(x + h) − f(x)] / h

Now take f(x) = x². The chord gradient becomes [(x + h)² − x²] / h = (2xh + h²) / h = 2x + h.

That is the whole derivation. For every non-zero h, the chord gradient is exactly 2x + h. As h shrinks, that value approaches 2x — and 2x is what we call the derivative.

The subtlety worth noticing: you cannot simply set h = 0, because the original fraction becomes 0/0, which means nothing. The limit language exists precisely to say "what this approaches" without ever dividing by zero.

Every standard derivative you will memorise was produced this way once.

02Why does the chain rule work?Because rates multiply.

If y depends on u, and u depends on x, then a small change in x produces a change in u, which produces a change in y.

If u changes three times as fast as x, and y changes five times as fast as u, then y changes fifteen times as fast as x. That is the entire idea:

dy/dx = (dy/du) × (du/dx)

The usual justification writes δy/δx = (δy/δu)(δu/δx) and cancels, which is honest arithmetic as long as δu ≠ 0.

Being straight with you: that condition can fail — u might be momentarily constant — and a rigorous proof handles that case separately. At A-level the multiplication-of-rates picture is the right one to carry, and it is genuinely why the rule is true.

It also tells you why the rule is not dy/du + du/dx: rates compose by scaling, not by stacking.

03Why is completing the square the whole of quadratics?Because it says every quadratic is a stretched, shifted copy of the same curve.

Any quadratic can be written a(x + p)² + q. Read that off and you have the vertex at (−p, q), the line of symmetry, and whether the curve opens up or down — with no calculation.

It also produces the formula. Start with ax² + bx + c = 0 and complete the square in general:

(x + b/2a)² = (b² − 4ac) / 4a²

Take the square root of both sides and rearrange, and the quadratic formula falls out. It is not a separate fact to memorise — it is completing the square carried out once, in general, so that nobody has to do it again.

The discriminant appears in the same line. b² − 4ac is under a square root, so its sign decides whether there are two roots, one, or none. That is why the criterion is that expression and not something else.

Three results — vertex, formula, discriminant — are one idea seen from three angles.

04Why do logarithms turn multiplication into addition?Because a logarithm is an index, and indices add.

The index law a^m × aⁿ = a^m+n says: to multiply powers of the same base, add the indices.

A logarithm asks the reverse question. log_a M means "what index turns a into M".

So if M = a^m and N = aⁿ, then MN = a^m+n, and reading the indices back gives log(MN) = log M + log N.

That is the whole law. It is the index law with the roles of the numbers and the indices swapped.

This also explains why log(M + N) does nothing useful: there is no index law for adding powers, so there is nothing for the logarithm to reverse.

For three hundred years this property was the fastest way to multiply large numbers. Slide rules are this law made physical.

05Why does antidifferentiating give you an area?Because the area accumulated so far is a function of where you stop — and its rate of growth is the height of the curve.

Let A(x) be the area under a curve from a fixed start to a movable point x. It is a function: move x, and the area changes.

Push x along by a small amount h. The extra area is a thin sliver, and if h is small the sliver is very nearly a rectangle of width h and height f(x). So the extra area is about f(x) × h.

[A(x + h) − A(x)] / h ≈ f(x)

But the left-hand side is exactly the chord-gradient expression from the derivative. Let h shrink and it becomes A′(x) = f(x).

So the area function is an antiderivative of the curve. That is why you find an antiderivative and evaluate it at the two ends: you are asking how much the area function grew between them.

This is the Fundamental Theorem of Calculus, and it deserves the name. Two ideas that look unrelated — the gradient of a tangent and the area under a curve — turn out to be inverse operations.

If one result from A-level is worth understanding rather than using, it is this one.

06Why does the scalar product tell you the angle?Because it is the cosine rule, rearranged.

Take two vectors a and b with angle θ between them. The vector from the tip of b to the tip of a is a − b, and those three vectors form a triangle.

The cosine rule on that triangle gives |a − b|² = |a|² + |b|² − 2|a||b| cos θ.

Now expand the left-hand side using the scalar product: (a − b)·(a − b) = |a|² − 2a·b + |b|².

Compare the two lines. Everything cancels except a·b = |a||b| cos θ.

The perpendicular case then needs no separate rule: if θ = 90° then cos θ = 0, so the scalar product is zero. That is why "dot product zero" and "perpendicular" are the same statement.

A geometric fact about triangles, wearing algebraic clothes.

07Why does the binomial expansion have those coefficients?Because they are counting something.

Expanding (x + y)ⁿ means multiplying out n identical brackets. From each bracket you take either an x or a y.

Every term in the answer comes from one such set of choices. A term with y^r comes from choosing y in exactly r of the brackets and x in the other n − r.

So the coefficient of x^n−ry^r is simply the number of ways of choosing which r brackets supply the y — which is ⁿC_r.

Pascal’s triangle is then not a curiosity. Each entry is the sum of the two above it because to choose r things from n, you either take the new item (and need r − 1 from the rest) or you do not (and need r from the rest).

Combinatorics and algebra turn out to be describing the same object.

The derivative, watched happening

The definition says a derivative is what the chord gradient approaches. This is that sentence, made visible. Shrink the gap and watch the number settle.

Drag the slider, or focus it and use the arrow keys.

Gap between the two points, h 1.00

Chord gradient — between x=1 and x=1+h

Exactly — 2x + h, with x = 1

The chord gradient is exactly 2 + h for every non-zero h. It never equals 2 — but it gets arbitrarily close, and 2 is the number it is closing in on.

The curve is y = x². The gold line joins (1, 1) to (1 + h, (1 + h)²). As h shrinks the two points converge and the chord becomes the tangent — which is what dy/dx = 2x is recording. You can never set h = 0: the fraction would be 0/0, and the two points would be the same point, which does not define a line at all.

How the topics join up

Taught as eleven separate units, A-level looks like eleven things to remember. It is closer to four or five ideas wearing different clothes.

Differentiation

Gradient of a tangent
Rate of change
Stationary points
Optimisation
Marginal cost and revenue

Marginal anything, in economics, is a derivative. When a textbook says "marginal cost is the cost of one more unit", it means the derivative of total cost.

Economics beyond the syllabus

Integration

Reverse of differentiating
Area under a curve
Accumulated total
Differential equations
Surplus in a market

Consumer surplus is the area between a demand curve and the price — which is to say, an integral.

How money moves

Sequences and series

Common ratio
Geometric series
Compound growth
Present value
Long-run totals

A geometric series is compound interest written as a sum. The same formula values a stream of payments and totals a bouncing ball.

Compound growth, modelled The time value of money

Exponentials and logarithms

Index laws
Exponential growth
The number e
Natural logarithms
Linearising a model

Taking logs of a curved relationship often straightens it. That is why so much economic data is plotted on a log scale — a straight line then means a constant growth rate.

Growth and the economics of AI

Probability and distributions

Counting outcomes
Random variables
Binomial and normal models
Expected value
Hypothesis testing

Expected value is the bridge from probability to decision-making, and the reason an insurer can price a risk it cannot predict.

What an actuary actually does Risk and return

Vectors

Directed quantities
Magnitude and direction
Scalar product
Geometry without coordinates
Forces in equilibrium

Mechanics is vectors with units. University linear algebra is vectors with the geometry removed and the structure kept.

Where vectors lead next

Stop asking which formula

Routine questions tell you the method in the wording. Hard ones do not, which is the entire difficulty. These are the approaches worth having available — each with a worked case small enough to check in your head.

What is actually being asked? Write the target in your own words before anything else.
What am I given, and what is it for? Unused information in a question is usually a signal you have missed something.
What structure do I recognise? A quadratic in disguise, a symmetry, a repeated block.
What assumptions are in play? Especially in mechanics and modelling, where they are marked.
Would another representation help? Graph, diagram, table, algebra.
Can I do a smaller version first?
Can I start from the answer and work back?
Does the size of my answer make sense?

The honest version

Most people, stuck, reread the question hoping to spot a keyword that maps to a method. That works on routine questions and fails completely on the ones that decide grades. The list on the left is what to do instead — and it is a skill that improves with deliberate practice, not a talent you either have or do not.

Try a small case

Before attacking the general problem, do the smallest version by hand. Patterns and structure that are invisible in general are often obvious for n = 1, 2, 3.

For instance. Asked to find the sum of the first n odd numbers, compute a few: 1, 4, 9, 16. They are squares. Now you know what to prove, which is a much easier position than not knowing what is true.

Reach for it when. Sequences, series, anything with a general n, and any "show that" where the target looks unmotivated.

Work backwards from the target

In a "show that" question, the answer is given. Start at the end and ask what would have to be true immediately before it.

For instance. To show tan θ + cot θ = 2 cosec 2θ, look at the target: it has a single trigonometric function of 2θ. So the last step almost certainly used a double-angle identity — which tells you what to aim the left-hand side at.

Reach for it when. Proofs and identities, where forwards is a search and backwards is a plan.

Change the representation

Algebra, graph, table, diagram and words are five views of the same object. Difficulty usually lives in one of them and not the others.

For instance. Solve |x − 1| < |x + 3|. Algebraically it is a case analysis. Read as distance — "which points are closer to 1 than to −3" — the answer is everything to the right of the midpoint, x > −1, in one line.

Reach for it when. Modulus, inequalities, and any question where the algebra is growing faster than the insight.

Look for structure before calculating

Spend ten seconds asking what kind of object this is before touching it. Symmetry, factors and special forms often remove most of the work.

For instance. ∫₋₂² x³ cos x dx looks like integration by parts twice. But the integrand is odd and the limits are symmetric, so the integral is zero. No working required.

Reach for it when. Definite integrals, series, and anything with symmetric limits or repeated structure.

Substitute to simplify

If an expression repeats, name it. Giving a messy chunk a single letter often turns an unfamiliar problem into one you have already met.

For instance. Solve 4^x − 5(2^x) + 4 = 0. Let u = 2^x. Since 4^x = (2^x)² = u², it is just u² − 5u + 4 = 0.

Reach for it when. Disguised quadratics, integration, and any expression where the same block appears twice.

Test an extreme case

Push a variable to zero, to one, or to infinity and check the answer still behaves. It will not prove anything, but it catches errors fast.

For instance. If a model gives the height of a projectile and setting t = 0 does not return the launch height, the algebra is wrong and you have found out in five seconds.

Reach for it when. After any long derivation, and before writing down a final answer to a modelling question.

Bound it before you solve it

Decide roughly what size the answer should be. A rough bound turns "is this right?" into a question you can answer.

For instance. An area between a curve and a chord must be smaller than the enclosing rectangle. If your integral exceeds it, you have a sign or a limit wrong.

Reach for it when. Areas, volumes, probabilities — anything with a natural ceiling. A probability outside 0 to 1 is a check you should never fail.

Find a counterexample

To disprove a general claim you need exactly one case where it fails. Looking for one is often quicker than trying to prove something false.

For instance. "If n² is even then n is even" is true. "If n² > n then n > 1" is not — take n = −2. Negative numbers and zero and one break more claims than anything else.

Reach for it when. Any "is it always true" question. Try 0, 1, −1 and a fraction before believing a statement.

Now try one

Knowing how to solve a problem is a different skill from knowing where to start. Twelve problems, chosen so that you already know all the mathematics involved and still have to decide what to do with it. Nothing here needs content beyond A-level.

Starting one idea, but not the first one you reach for Think a decision about method or representation Challenge an observation is needed before any calculation Deep synthesis, or an argument rather than an answer

Divide by x at your peril

Inequalities · domain

STARTING

Solve x² > x.

Hint

Whatever you do to both sides, ask whether the thing you did to them could have been negative.

Solution

The instinct is to divide both sides by x, which gives x > 1. That is wrong, and it is wrong twice over: dividing by x assumes x is not zero, and it assumes x is positive, because dividing an inequality by a negative number reverses it.

Instead move everything to one side and factorise:

x² − x > 0 ⟹ x(x − 1) > 0

Now the question is: when is a product of two things positive? When both factors are positive, or both are negative. Both positive needs x > 1. Both negative needs x < 0.

So the solution is x < 0 or x > 1. Testing x = −2 confirms it: 4 > −2.

Why this works

Multiplying or dividing an inequality by an expression whose sign you do not know is not a legal move — it is three different moves depending on that sign, and you have to handle all three. Rearranging to compare with zero avoids the problem entirely, because a product being positive or negative is a statement about signs rather than sizes. This is the same habit that makes quadratic inequalities and rational inequalities routine later.

The tempting wrong turn

Dividing by x and answering x > 1. It is tempting because it is one step and it produces something that looks like an answer — and it is even half right. The tell is that you divided by a letter without saying anything about its sign. If you ever find yourself doing that, stop.

Go further

Now solve x³ > x. The same method works, but the factorisation is x(x − 1)(x + 1) > 0 and there are now three sign changes to track — try a number line marked at −1, 0 and 1 and check one value in each of the four regions.

Algebraic manipulation Why completing the square is the whole of quadratics

The curve with something missing

Functions · domain

STARTING

Two students are asked to sketch y = (x² − 1)/(x − 1). One draws the line y = x + 1. The other says that is wrong. Who is right?

Hint

Cancel the fraction, then ask what you were allowed to assume in order to cancel.

Solution

The numerator factorises: x² − 1 = (x − 1)(x + 1). Cancelling gives y = x + 1, so the first student looks right.

But cancelling (x − 1) from top and bottom means dividing by x − 1, and that is only legal when x ≠ 1. At x = 1 the original expression is 0/0, which is not a number.

So the graph is the line y = x + 1 with the single point (1, 2) removed — usually drawn as a small open circle.

The second student is right, though only just: the two expressions agree at every value of x except one.

Why this works

A function is not only a formula — it is a formula together with a domain. Two rules that produce the same output everywhere they are both defined are still different functions if their domains differ. This distinction looks pedantic here and stops looking pedantic the moment you meet limits, because the whole point of a limit is to describe what a function does near a point where it may not be defined at that point. The derivative is exactly such a case.

The tempting wrong turn

Cancelling without comment and drawing an unbroken line. It is tempting because the algebra is correct — the error is not in the cancelling but in forgetting to record what the cancelling assumed. Watch for any step where you divide by something containing the variable.

Go further

What does the graph of y = (x² − 1)/(x + 1) look like, and where is the hole this time? Then consider y = (x − 1)/(x² − 1), which behaves quite differently — one of the two problem points becomes an asymptote rather than a hole. Deciding which is which is the useful skill.

Why differentiation gives a gradient Algebra and functions

Do not solve for x

Algebraic structure

THINK

Given that x + 1/x = 3, find x² + 1/x².

Hint

You are not asked for x. What happens if you square the thing you were given?

Solution

The obvious route is to solve for x. Multiplying through gives x² − 3x + 1 = 0, so x = (3 ± √5)/2, and then you square that, invert it, and add. It works. It is also several minutes of unpleasant surd arithmetic.

The observation that removes all of it: square the equation you were handed.

(x + 1/x)² = x² + 2·x·(1/x) + 1/x² = x² + 2 + 1/x²

The cross-term is 2x × 1/x = 2 — the x cancels, which is the whole reason this works.

So 3² = x² + 1/x² + 2, giving x² + 1/x² = 7.

Why this works

The question asks about a symmetric expression — one unchanged if you swap x for 1/x. Symmetric expressions in x and 1/x can always be built from x + 1/x without ever knowing x, because the cross-terms cancel every time. Noticing the structure of what is being asked for, before deciding what to do, is what turns a five-minute problem into a five-second one.

The tempting wrong turn

Solving the quadratic. It is not a mistake — it reaches the right answer — but it is choosing the familiar method over the one the question is shaped for. That instinct is exactly what makes harder papers feel long. The tell: you are computing something the question never asked for.

Go further

Find x³ + 1/x³ from the same starting point. Cubing gives (x + 1/x)³ = x³ + 1/x³ + 3(x + 1/x), so 27 = X + 9 and X = 18. Then ask the sharper question: is there any real x with x + 1/x = 1?

Look for structure before calculating Why the binomial coefficients are what they are

One equation, two unknowns, one answer

Completing the square

THINK

Find all real numbers x and y satisfying x² + y² = 4x − 6y − 13.

Hint

Get everything onto one side. Then deal with the x terms and the y terms separately, in the way you would if each were alone.

Solution

One equation and two unknowns usually means infinitely many solutions, so the phrase "find all" is a signal that something unusual is happening here.

Move everything left: x² − 4x + y² + 6y + 13 = 0.

Complete the square in each variable separately. x² − 4x = (x − 2)² − 4 and y² + 6y = (y + 3)² − 9.

(x − 2)² − 4 + (y + 3)² − 9 + 13 = 0

The constants are −4 − 9 + 13 = 0, so this collapses to (x − 2)² + (y + 3)² = 0.

Two squares of real numbers, adding to zero. Neither can be negative, so neither can be positive either — each must be exactly zero. Hence x = 2 and y = −3, and that is the only solution.

Why this works

Completing the square converts an equation about sizes into an equation about squares, and squares carry a sign restriction for free: they are never negative. Whenever a sum of squares is forced to equal zero, every term is pinned individually. The same equation with 12 instead of 13 on the right would give a circle of radius 1 — so this is really the degenerate member of the family of circles, the one whose radius has shrunk to nothing.

The tempting wrong turn

Concluding that one equation in two unknowns cannot pin down both, and stopping. It is a reasonable rule of thumb and it is usually right — but it counts equations rather than looking at them, and the constraint here comes from the squares rather than from the count.

Go further

Replace 13 with a general constant k. For which k does the equation describe a circle, for which a single point, and for which nothing at all? You will find the boundary case is exactly the one above, and the criterion behaves like a discriminant.

Why completing the square is the whole of quadratics Coordinate geometry

Can you avoid differentiating?

Optimisation · inequalities

THINK

For x > 0, find the smallest possible value of x + 1/x — and then prove your answer is right without using calculus.

Hint

For the second part: you want to show the expression is never below some number. What kind of quantity is never negative?

Solution

With calculus it is quick. d/dx (x + x⁻¹) = 1 − 1/x², which is zero when x² = 1. Since x > 0 we take x = 1, giving the value 2.

Without calculus, the useful idea is that a square is never negative. Consider (√x − 1/√x)², which is defined because x > 0. Expanding:

(√x − 1/√x)² = x − 2 + 1/x

That is exactly the expression we care about, minus 2. And the left-hand side is a square, so it is at least zero:

x + 1/x − 2 ≥ 0, so x + 1/x ≥ 2

Equality needs the square to be zero, which needs √x = 1/√x, so x = 1. The minimum is 2, attained only at x = 1.

Why this works

The two methods answer different questions. Calculus finds where the minimum is by locating a flat point; the algebraic argument shows why 2 is a floor, because it rewrites the whole expression as "2 plus something that cannot be negative". The second is stronger: it establishes the bound for every x at once, rather than checking a candidate. This is a small instance of a general pattern — inequalities are often proved by manufacturing a square.

The tempting wrong turn

Reaching for the derivative and stopping there. It is not wrong, but it leaves you unable to answer "why 2 and not something else", and it needs a second-derivative check to confirm the stationary point is a minimum at all. The other trap is dropping the condition x > 0: for negative x the expression has a maximum of −2, and the unrestricted function has no minimum whatsoever.

Go further

The inequality x + 1/x ≥ 2 is the two-term case of the arithmetic–geometric mean inequality, which says the average of a set of positive numbers is never below their geometric mean. Try proving a + b ≥ 2√(ab) for positive a and b by the same square trick.

Look for structure before calculating Differentiation

Do not expand it first

Calculus · optimisation

THINK

An open box is made from a square sheet of side a by cutting a square of side x from each corner and folding up the flaps. Show that the volume is greatest when x = a/6.

Hint

Write the volume in factorised form and differentiate it that way. Resist multiplying out.

Solution

The base is a square of side a − 2x and the height is x, so V = x(a − 2x)², valid for 0 < x < a/2.

Differentiate as a product, keeping the bracket intact:

dV/dx = (a − 2x)² + x · 2(a − 2x)(−2)

Both terms contain (a − 2x), so take it out:

dV/dx = (a − 2x)[(a − 2x) − 4x] = (a − 2x)(a − 6x)

The stationary points are therefore x = a/2 and x = a/6, read straight off with no quadratic formula.

At x = a/2 the base has shrunk to nothing and V = 0 — it is the endpoint of the domain, not a maximum. For x slightly below a/6 both factors are positive so V is increasing; just above, the second factor turns negative so V is decreasing. So x = a/6 is the maximum.

Why this works

Differentiating a product in factorised form leaves a common factor sitting in both terms, and factorising it out hands you the roots directly. Expanding first gives V = a²x − 4ax² + 4x³ and a derivative you then have to factorise again — the same work, done twice, with an extra opportunity for a sign error. The general habit: the form an expression is already in is usually the form its derivative wants to stay in.

The tempting wrong turn

Expanding, differentiating, then solving 12x² − 8ax + a² = 0 with the quadratic formula. It gets there. The subtler error is finding both roots and declaring x = a/6 the maximum without saying why x = a/2 is rejected — the reason is the physical domain, not the calculus.

Go further

What if the sheet is a rectangle a × b rather than a square? The derivative is still a quadratic in x, but now the answer genuinely needs the formula, and only one of the two roots lies in the valid range. Working out which, and why, is the interesting part.

Differentiation Look for structure before calculating

A sequence with a secret

Sequences · recurrence

CHALLENGE

A sequence is defined by u₁ = 1 and u_n+1 = u_n / (1 + u_n). Find a formula for u_n.

Hint

Write out the first four terms. Then, if the pattern is not obvious, try looking at the reciprocals instead.

Solution

Compute a few terms. u₁ = 1; u₂ = 1/2; u₃ = (1/2)/(3/2) = 1/3; u₄ = (1/3)/(4/3) = 1/4.

So it looks like u_n = 1/n. Guessing is legitimate — but a guess is not a result, so check it against the recurrence. If u_n = 1/n then

u_n+1 = (1/n)/(1 + 1/n) = (1/n)/((n+1)/n) = 1/(n+1)

which is the formula again with n replaced by n+1. Together with u₁ = 1 that settles it.

There is a better route that finds the answer rather than confirming it. Take reciprocals of the recurrence:

1/u_n+1 = (1 + u_n)/u_n = 1/u_n + 1

Writing v_n = 1/u_n, this says v_n+1 = v_n + 1 with v₁ = 1 — an arithmetic sequence with common difference 1. So v_n = n and u_n = 1/n.

Why this works

The recurrence is awkward because the unknown appears in both the numerator and the denominator. Taking reciprocals moves it to one place and the messy relation becomes the simplest possible one. This is changing the representation: the sequence was never complicated, it was written in unhelpful coordinates. The same manoeuvre — transform, solve in the easy setting, transform back — is behind logarithms, substitution in integration, and a great deal of university mathematics.

The tempting wrong turn

Computing three terms, spotting 1/n, and writing it down as the answer with no verification. The pattern is real here, so nothing goes wrong — which is exactly what makes the habit dangerous. Problem 11 is the same habit meeting a sequence that betrays it.

Go further

Try u_n+1 = u_n/(1 + 2u_n) with u₁ = 1. The reciprocal trick still works, and the resulting arithmetic sequence has a different common difference. What is the general pattern for u_n+1 = u_n/(1 + ku_n)?

Change the representation Sequences and series

The test that is usually wrong

Conditional probability

CHALLENGE

A condition affects 1 person in 1000. A test for it is 99% accurate in both directions: 99% of people who have it test positive, and 99% of people who do not test negative. You test positive. What is the probability that you have the condition?

Hint

Do not reason with percentages. Take a population of 100,000 people and count how many end up in each of the four possible boxes.

Solution

Take 100,000 people. Of these, 100 have the condition and 99,900 do not.

Of the 100 who have it, 99% test positive: 99 true positives.

Of the 99,900 who do not, 1% test positive anyway: 999 false positives.

So the number of positive results altogether is 99 + 999 = 1098, and only 99 of those people actually have the condition.

P(condition | positive) = 99 / 1098 ≈ 0.090

About 9%. Roughly ten positive results in eleven are wrong, from a test that is right 99% of the time.

Why this works

The 99% figure answers a different question from the one you asked. It is P(positive | condition) — the probability of the evidence given the hypothesis. What you want is P(condition | positive), the hypothesis given the evidence. These are not the same number and can differ enormously, because the second depends on how common the condition is in the first place. The condition is so rare that the small percentage of false positives is drawn from a vastly larger group, and it swamps the true positives.

The tempting wrong turn

Answering 99%. It is overwhelmingly the common response, including among people who use tests professionally, and the reason is that the two conditional probabilities read almost identically in English. The tell is the phrase "the test is 99% accurate", which never tells you what you want on its own — you also need the base rate.

Go further

Rework it with the condition affecting 1 person in 20 instead of 1 in 1000, and the answer becomes about 84%. The test has not changed at all. This is why screening a whole population and testing people with symptoms give such different results, and it is the arithmetic behind Bayes and the deeper probability work. The same reasoning governs how insurers price a risk they cannot predict for any individual.

Probability and statistics What an actuary does with this

What perpendicular actually means

Vectors

CHALLENGE

Prove that the diagonals of a rhombus are perpendicular. Use vectors, and do not set up coordinates.

Hint

Call two adjacent sides a and b. What are the two diagonals in terms of those? What does a rhombus tell you about a and b?

Solution

Let two adjacent sides of the rhombus be the vectors a and b, starting from the same vertex. A rhombus is a parallelogram with all sides equal, so |a| = |b|.

Going along one side then the other reaches the opposite corner, so one diagonal is a + b. The other diagonal joins the tips of a and b, and is a − b.

Two vectors are perpendicular exactly when their scalar product is zero, so compute it:

(a + b)·(a − b) = a·a − a·b + b·a − b·b

The scalar product is commutative, so the two middle terms cancel. What remains is |a|² − |b|².

Since the sides are equal, that is zero. The diagonals are perpendicular.

Note what the proof also tells you: it works only because the sides are equal. In a general parallelogram |a| ≠ |b| and the diagonals are not perpendicular — the algebra fails at exactly the point where the geometry does.

Why this works

This is the difference-of-two-squares identity, living in vectors. The scalar product distributes over addition and is commutative, so (a+b)·(a−b) expands exactly like (p+q)(p−q) — and that is why the cross-terms disappear. Coordinates would have worked too, but they would have buried the reason under arithmetic. The proof is short precisely because vectors carry the geometric condition (equal lengths) directly into the algebra.

The tempting wrong turn

Placing the rhombus on axes with a vertex at the origin, writing down four coordinates, and computing two gradients. It is a valid proof, but it takes far longer, it needs care to keep the rhombus general rather than a special case, and at the end you have verified the fact without seeing what caused it.

Go further

Use the same method on the converse: if the diagonals of a parallelogram are perpendicular, must it be a rhombus? The algebra runs backwards cleanly, which is worth noticing — many geometric statements are not reversible, and checking which direction a proof actually establishes is a habit worth having.

Why the scalar product tells you the angle Vectors

Can this model be trusted?

Exponential modelling

CHALLENGE

A population is modelled by P = 500e^0.08t, with t in years. Find the doubling time. Then use the model to predict the population after 200 years, and say what your answer tells you about the model.

Hint

The second part is not really a calculation question. Work out the number, then ask whether you believe it.

Solution

For the doubling time, solve e^0.08t = 2. Taking natural logarithms, 0.08t = ln 2, so t = ln 2 / 0.08 ≈ 8.66 years.

Notice this does not depend on the 500 at all: exponential growth doubles in the same time from any starting point, which is what distinguishes it from linear growth.

After 200 years, P = 500e¹⁶. Since e¹⁶ ≈ 8.9 × 10⁶, this gives roughly 4.4 × 10⁹ — about 4.4 billion, from a starting population of 500.

That is the answer, and it is the point. No real population behaves like this: food, space and disease impose limits that the model contains no representation of. The model is not broken — it is being used outside the range where its assumptions hold.

A defensible response is that the model is reasonable while the population is small relative to the resources available, and useless once it is not.

Why this works

Every model is a set of assumptions wearing an equation. P = P₀e^kt is the solution of dP/dt = kP, which says the growth rate is proportional to the current size and nothing else — no ceiling anywhere in the statement. Unbounded growth is therefore not a flaw in the arithmetic, it is a faithful consequence of what was assumed. Recognising which of your results are consequences of reality and which are consequences of your assumptions is most of what modelling marks are actually for.

The tempting wrong turn

Computing 4.4 billion and writing it down as the prediction. The calculation is right; the failure is treating the model as a machine that produces facts rather than as an argument with conditions attached. A useful reflex: whenever a model is extrapolated far beyond its data, check whether the answer is physically possible before you present it.

Go further

The standard repair is the logistic model, dP/dt = kP(1 − P/M), where M is a carrying capacity. Without solving it, read the equation: what happens to the growth rate when P is small, and what happens as P approaches M? Being able to interpret a differential equation you cannot solve is a genuinely useful skill.

Exponentials and logarithms Differential equations

Always true?

Proof · counterexample

DEEP

A student claims: for every positive integer n, the value of n² + n + 41 is prime. They have checked n = 1 to n = 10 and it worked every time. Is the claim true?

Hint

Testing more values is one option, and it will take a while. Instead, try to make the expression factorise — is there a value of n that would obviously spoil it?

Solution

The claim survives a lot of testing. For n = 1 it gives 43, then 47, 53, 61, 71, and on it goes — every value from n = 1 to n = 39 is prime. That is far more evidence than most people would demand.

It is false anyway. Rather than testing further, look for a value of n that forces a factor. If n = 41, then every term is a multiple of 41:

41² + 41 + 41 = 41(41 + 1 + 1) = 41 × 43

which is not prime. In fact the claim fails one step earlier, at n = 40:

40² + 40 + 41 = 1600 + 81 = 1681 = 41²

So n = 40 is the first counterexample, and one counterexample is enough. The claim is false.

Why this works

A universal statement — "for every n" — makes infinitely many assertions, and checking finitely many of them can never establish it. But its negation is an existence claim, and existence claims are settled by producing one example. That asymmetry is why disproof is often far easier than proof, and why hunting for a counterexample is a legitimate first move rather than an admission of defeat. Notice also how the counterexample was found: not by searching, but by asking what would make the expression factorise. Structure, again, rather than brute force.

The tempting wrong turn

Testing to n = 10, or 20, or 30, and concluding the claim is true. This polynomial is famous precisely because it punishes that reasoning so patiently — thirty-nine consecutive successes, and then failure. Evidence and proof are different categories, and no quantity of the first becomes the second.

Go further

Show that no polynomial with integer coefficients can produce a prime for every positive integer n. The argument is within reach: if f(1) = p is prime, consider f(1 + kp) for whole numbers k and show that p divides it. That rules out the whole strategy at once, rather than one polynomial at a time.

Proof and mathematical argument Find a counterexample Where proof is the whole game

Why the area under 1/x is a logarithm

Integration · logarithms

DEEP

Let A(t) be the area under y = 1/x from x = 1 to x = t, for t > 1. Show that A(ab) = A(a) + A(b) — without using the fact that the integral of 1/x is ln x.

Hint

Split the area from 1 to ab at the point a. Then try to turn the second piece into an area that starts at 1, using a substitution that stretches the axis.

Solution

Split the region at x = a:

A(ab) = ∫₁^a dx/x + ∫_a^ab dx/x = A(a) + ∫_a^ab dx/x

So everything rests on showing the second piece equals A(b). Substitute x = au, so dx = a du. When x = a, u = 1; when x = ab, u = b.

∫_a^ab dx/x = ∫₁^b (a du)/(au) = ∫₁^b du/u = A(b)

The factor a from dx cancels the a in the denominator exactly. That cancellation is the entire content of the result.

Therefore A(ab) = A(a) + A(b).

Why this works

You have just derived the law of logarithms from an area, without ever mentioning logarithms. That is the striking part: the function "area under 1/x" turns products into sums purely because of the shape of the curve, and turning products into sums is the defining property of a logarithm. The cancellation works because 1/x is the one power for which stretching the horizontal axis by a factor squashes the height by exactly the same factor, leaving the area unchanged. This is why the natural logarithm can be defined as this area, and why 1/x is the gap in the pattern ∫xⁿ dx = xⁿ⁺¹/(n+1): at n = −1 that formula divides by zero, and something of a different kind has to take over.

The tempting wrong turn

Writing A(t) = ln t immediately and citing ln(ab) = ln a + ln b. It is true and it is circular — the question is asking you to establish the property, so assuming the function that has it settles nothing. Recognising when an argument has quietly assumed its conclusion is a large part of what proof questions test.

Go further

Use the same substitution to show A(tⁿ) = n A(t) for positive integers n, which is the power law. Then ask what A does for t between 0 and 1, and why the area comes out negative — and what that says about ln of a number below 1.

Why logarithms turn multiplication into addition Why antidifferentiating gives an area

Your mistakes are information

Writing the correct answer in red next to a wrong one teaches you almost nothing. Which kind of mistake it was tells you what to do differently — and the five kinds need five different responses.

Conceptual

“I do not actually understand what this is.”

How you know. You can follow a worked solution line by line but could not have started it, and cannot say what the answer means.

What to change. Stop doing questions. Go back to what the object is — read the "why it works" section for that topic, and try to explain it aloud without notation.

Procedural

“I knew the method and made a mess of it.”

How you know. Your method line is right and the arithmetic or algebra after it is not. Sign errors, dropped factors, mis-expanded brackets.

What to change. The cure is not more questions, it is slower ones. Most procedural errors cluster — find out whether yours are signs, indices or fractions, then drill that one thing.

Representational

“I could not turn the words into mathematics.”

How you know. You stall before writing anything, or you write down the wrong quantity — modelling the distance when the question asked about the rate.

What to change. Practise the translation on its own: read a question and write only the variables, what is known, and what is asked. Do not solve it.

Strategic

“I did not know where to start.”

How you know. You recognise every technique in the solution once you see it, but nothing suggested itself. This is the commonest reason strong students stall.

What to change. This is the one the problem-solving approaches exist for. It is a skill, not a talent, and it responds to deliberately practising unfamiliar problems.

Checking

“I had it right and did not notice I had gone wrong.”

How you know. A probability above 1, a negative length, a speed that grows without limit, an answer that fails at t = 0.

What to change. Build one habit: before writing the final line, ask what size the answer should be and whether it behaves at the extremes.

Five questions worth asking about any topic

Not a progress bar and not a score. Just the questions that separate having met an idea from actually having it.

01
Can you state it?
Write down what the idea says, without notation, in a sentence someone in the year below would follow.
02
Can you use it without a template?
Solve a question that is not the same shape as the example you learned it from.
03
Can you say why it works?
Not the proof necessarily — the reason. If the only answer is "because that is the rule", you have not finished.
04
Can you connect it?
Name one other topic where the same idea appears. Most A-level ideas appear in at least two places.
05
Can you solve something unfamiliar with it?
A problem where nobody has told you which method to use. This is what the top grades are actually testing.

Where it goes next

Not harder A-level. Different mathematics — the ideas the syllabus points at and then leaves. None of this needs to be done in order, and none of it is required for an A*.

Proof by induction

A way to prove a statement for every positive integer by proving it for one, and proving that truth at each step forces truth at the next.

Why it is interesting: It is the first proof technique that feels like it should not work — you prove infinitely many statements by writing two. Getting comfortable with why it is legitimate changes how you think about the integers.
Builds on: Sequences and series, and the idea of a general term.
Then look at: Strong induction, and then recursive definitions and structural induction in computer science.

Complex numbers

What happens when you stop refusing to take the square root of a negative number and see what follows.

Why it is interesting: The astonishing part is that inventing one new number makes every polynomial of degree n have exactly n roots, and unifies exponentials with trigonometry. Nothing at A-level prepares you for how tidy it becomes.
Builds on: Quadratics with negative discriminant, and the trigonometric identities.
Then look at: The Argand diagram, de Moivre’s theorem, and roots of unity.

Matrices and linear algebra

Arrays of numbers that represent transformations — rotations, stretches, projections — and that can be composed by multiplying them.

Why it is interesting: It explains why matrix multiplication has its strange definition: it is composition of functions. Most of modern applied mathematics, including everything under machine learning, is linear algebra.
Builds on: Vectors, transformations of graphs, and simultaneous equations.
Then look at: Determinants, eigenvectors, and what it means for a transformation to have a direction it does not rotate.

Differential equations

Equations whose unknown is a function, described by how it changes rather than by what it is.

Why it is interesting: Almost every quantitative model of the real world is one. Populations, cooling, epidemics, interest and motion are all statements about rates, and the solution is the behaviour they force.
Builds on: Integration, and any question phrased as "the rate of change is proportional to…".
Then look at: Second-order equations, systems of equations, and when a closed-form solution does not exist.

Deeper probability

Conditional probability taken seriously — Bayes, independence, and why intuition fails so reliably here.

Why it is interesting: The medical-test paradox, where a positive result on a highly accurate test for a rare condition still usually means you are fine, is arithmetic that almost everybody gets wrong. Understanding it changes how you read statistics in the news.
Builds on: Conditional probability and tree diagrams.
Then look at: Bayes’ theorem, prior and posterior belief, and the statistical reasoning behind risk models.

Optimisation under constraint

Maximising something when you are not free to choose everything — the most useful shape of problem in applied mathematics.

Why it is interesting: A-level optimisation gives you a single variable and asks for a maximum. Real problems have several variables and a budget, which is a genuinely different and more interesting question.
Builds on: Stationary points, and the second derivative test.
Then look at: Lagrange multipliers, and the constrained-choice problems at the centre of microeconomics.

Competitions that use this kind of thinking What the TMUA actually asks for Economics past the syllabus

Mathematics, and what it is actually doing.

Getting better is not about knowing more.

MathematicsYou are here

Economics

I have just finished GCSEs

I am starting A-level Maths

I know the content but hard questions defeat me

I want to go past the syllabus

I am aiming at TMUA or olympiad problems

Proof and mathematical argument

Algebra and functions

Coordinate geometry

Sequences and series

Trigonometry

Exponentials and logarithms

Differentiation

Integration

Vectors

Probability and statistics

Mechanics

Differentiation

Integration

Sequences and series

Exponentials and logarithms

Probability and distributions

Vectors

Try a small case

Work backwards from the target

Change the representation

Look for structure before calculating

Substitute to simplify

Test an extreme case

Bound it before you solve it

Find a counterexample

Divide by x at your peril

The curve with something missing

Do not solve for x

One equation, two unknowns, one answer

Can you avoid differentiating?

Do not expand it first

A sequence with a secret

The test that is usually wrong

What perpendicular actually means

Can this model be trusted?

Always true?

Why the area under 1/x is a logarithm

Can you state it?

Can you use it without a template?

Can you say why it works?

Can you connect it?

Can you solve something unfamiliar with it?

Proof by induction

Complex numbers

Matrices and linear algebra

Differential equations

Deeper probability

Optimisation under constraint