Everything, on one sheet

The card

Structure: assume the OPPOSITE. Derive correctly. State WHY what you reached is impossible. Conclude the assumption is false, so the original is true.
M1 (real Jun 2022 MS): the opening mark is generous on wording — a correct equation earns it, no fixed phrase required.
The closing mark needs a REASON the contradiction is impossible (even ≠ odd; a square can't be negative) — not just 'no solutions.'
Spec-named examples: √2 is irrational; there are infinitely many primes. Also applies to statements you've never seen before.
Contradiction proves a TRUE, unbounded 'for all' claim. Exhaustion needs a finite domain; counter-example needs a FALSE claim.
A case split inside a proof (e.g. n even / n odd) is itself a for-all claim — both branches must reach their own contradiction.
A target number's OWN divisors can turn an infinite search into a short, finite list to exhaust (e.g. 4p² − q² = 46 → (2p+q)(2p−q) = 46 has exactly 2 valid factor pairs) — exhaustion nested inside a contradiction. Every pair on the list must be checked, not just one.
A real Jan 2026 mark scheme credits three separate routes on equal footing for one question — divisor-pair testing, odd/even parity, and a q = 2m substitution: proof by contradiction often has more than one legitimate path to the same contradiction.

Why it works — Why one impossibility proves an infinite claim — and why a case split inside it still has to be exhaustive

This paper's WMA12 unit — read as this lesson's own prerequisite — derives a fact worth carrying over directly: a claim of the form "true for every n" is really an infinite AND, one statement per value of n, joined together. Proving that AND directly, case by case, is proof by exhaustion, and it only works when the list of cases is finite and short enough to write out. Contradiction sidesteps the infinite list a different way: instead of checking every case, it works with a single, general, unspecified case — an arbitrary integer n, or an arbitrary fraction p/q — that stands for every possible instance of the claim at once. If assuming that this one general case fails to satisfy P leads to something impossible, the argument never depended on which particular n, p or q was chosen — so no case, out of infinitely many, can be the exception. That is the entire reason contradiction can establish a TRUE claim over an infinite domain, which is precisely the one thing exhaustion (bounded to a finite list) and counter-example (which only ever disproves a FALSE claim) cannot do. There is one place this logic bites back inside a contradiction proof itself, and it is worth stating precisely because it is where marks are actually lost: if the derivation genuinely splits into more than one case — as it does below, where an assumed integer n is either even or odd — then "every integer is even or odd" is itself a two-case for-all claim, and by the exact same AND logic, BOTH branches have to be followed to their own impossibility before the proof is finished. Reaching a contradiction down one branch and treating the whole argument as settled is not a smaller version of a correct proof; it is an incomplete one, in exactly the way a real examiner report records candidates losing marks for reaching "the equation formed had no solutions" and stopping — the same incompleteness, whether it is a missing reason on one branch or a missing branch altogether.

Traps — 7

assumption-and-conclusion-omitted

The paper's own general commentary states this plainly: candidates "often omit questions on this topic or struggle to adopt a suitable strategy to complete the proof" (Jan 2021, general report), and a specific question the same series records the shape of that struggle directly: "The majority obtained the method mark for suggesting two appropriate odd numbers but full proofs with assumption, reason and conclusion were less common" (Jan 2021, Q3). On this evidence, the algebra in the middle is not where most marks are actually lost — the sentence before it and the sentence after it are.

plausible-but-false-assertion-in-place-of-derivation

Confirmed directly, and worth reading twice: "False reasoning was sometimes seen, for example: 'n is an integer so n² + 1 is odd'" (Jan 2021, Q3). Read as a claim about every integer n, this sentence is not merely unjustified — it is false: n² + 1 is odd exactly when n is even (n = 2 gives 5) and even when n is odd (n = 3 gives 10). A sentence that reads like a derivation but states something untrue is a worse failure than a visible gap in the working, precisely because it looks finished.

logical-step-skipped-before-the-conclusion-that-needs-it

On a two-part proof building toward the irrationality of √2: "A great many responses to part (a) did correctly factorise the expression, but few made a comment to state that it was odd. Both of these aspects were required to show the contradiction" (Oct 2021, Q10). The pattern here is an ordering failure, not an arithmetic one — reaching a correct intermediate expression and moving straight to what follows from it, without writing the sentence that actually licenses the move. This is exactly the gap the worked chain above forces open at stage 3 and stage 5, where "p is even" and "q is even" are each derived with their own stated reason, not asserted by analogy with each other.

lowest-terms-condition-not-stated

Confirmed on the same question, and specific to any proof built on a fraction assumed to be in simplest form: "Most of these forgot, however, to add a statement that a/b was fully simplified, which would mean that the last mark in the question could not be awarded. It is really important in a proof to include all necessary steps" (Oct 2021, Q10). This is the exact reason the √2 chain above states the lowest-terms condition in its first line rather than its last — it is needed again at the end, to say precisely what the final contradiction (both p and q even) actually contradicts.

impossible-claim-asserted-without-a-reason

On a proof by contradiction that a cubic has no stationary points: "The majority successfully set up their initial assumption... However, many then simply commented that the equation formed had no solutions, giving no explanation as to why, or they gave an inadequate justification, and so gained no further credit" (Jan 2024, Q8). Reaching the right equation is not the same as explaining why it cannot hold — "this has no solutions" is an assertion; "the left side is even and the right side is odd" is the reason that assertion actually needs.

non-algebraic-check-substituted-for-a-required-contradiction

On the same question, naming the specific wrong tools candidates reached for instead of the required derivation: "attempting the 'discriminant', using a graph only, testing values of x or attempting to use small angle approximations" (Jan 2024, Q8) — none of which scored, because a proof by contradiction on this paper is assessed on an algebraic derivation of an impossibility, not on evidence that a check was carried out. A graph or a handful of tested values can make a statement look true; only a derived contradiction proves it.

not-all-valid-factor-pairs-checked

A real WMA14 mark scheme names this exact gap directly, for the one mark that checks it: the dependent mark on the divisor-pair route to Q8 requires "States and attempts to solve both valid pairs of equations" (Jan 2026, Q8) — not one of the two pairs 46×1 and 23×2, however correctly solved, but both. This is a different shape of incompleteness from the n-even/n-odd split earlier in this lesson: there, the two branches come from every integer's own parity; here, they come from every way one specific target number (46) actually factors. The underlying reason both fail is identical, and it is this lesson's own mechanism block that derives it: a proof that rules out only some of the possibilities the assumption allows for has not yet ruled out the assumption itself.

The card

a·b = a₁b₁+a₂b₂+a₃b₃ (NOT in the booklet — memorise). cosθ = a·b/(|a||b|). a·b=0 ⇔ a ⟂ b (a, b ≠ 0).
Line: r = a + tb (a = a point on the line, b = direction) or r = c + t(d − c), where d − c is the direction.
Parallel? Direction vectors must be scalar multiples — check EVERY component with the SAME k, not just some.
Not parallel: solve any 2 of the 3 component equations for the parameters, then check the 3rd. Consistent → intersecting (state the point). Inconsistent → skew.
Skew needs BOTH not-parallel AND not-intersecting shown — 'not perpendicular' proves neither.
Angle ABC at a named vertex: use BA and BC, report exactly what you get, obtuse or not. Angle BETWEEN two lines (no vertex): report the acute value, converting via 180°−θ if the raw result is obtuse.
Area of the parallelogram spanned by u, v from a shared vertex = |u||v|sinθ (triangle = half that); sinθ = √(1−cos²θ), reusing the same θ found from the scalar product — reversing either vector doesn't change the area, unlike the angle itself.
Cross product (NOT P4 spec content, but a real mark scheme credits it as a full-marks alternative): u×v = (u₂v₃−u₃v₂, u₃v₁−u₁v₃, u₁v₂−u₂v₁); |u×v| = |u||v|sinθ = the parallelogram's area directly, no angle needed.

Why it works — Where cos θ = a·b / (|a||b|) actually comes from — the cosine rule, not a new rule

Place the origin at O, so $\mathbf{a} = \overrightarrow{OA}$ and $\mathbf{b} = \overrightarrow{OB}$ are position vectors and $\theta = \angle AOB$. Triangle $OAB$ has sides $OA = |\mathbf{a}|$, $OB = |\mathbf{b}|$, and — by the 7.4 identity already used above — $AB = |\mathbf{b}-\mathbf{a}|$. The cosine rule, confirmed present in the formula booklet's P1 section (WMA14-verified-facts.md §3), states $AB^2 = OA^2 + OB^2 - 2\cdot OA\cdot OB\cos\theta$, i.e. $|\mathbf{b}-\mathbf{a}|^2 = |\mathbf{a}|^2+|\mathbf{b}|^2-2|\mathbf{a}||\mathbf{b}|\cos\theta$. Now expand the left side using the scalar product's own defining property, $|\mathbf{v}|^2 = \mathbf{v}\cdot\mathbf{v}$: $(\mathbf{b}-\mathbf{a})\cdot(\mathbf{b}-\mathbf{a}) = \mathbf{b}\cdot\mathbf{b} - 2\,\mathbf{a}\cdot\mathbf{b} + \mathbf{a}\cdot\mathbf{a} = |\mathbf{b}|^2 - 2\,\mathbf{a}\cdot\mathbf{b} + |\mathbf{a}|^2$. Setting the two expressions for $AB^2$ equal and cancelling the $|\mathbf{a}|^2+|\mathbf{b}|^2$ common to both sides leaves $-2\,\mathbf{a}\cdot\mathbf{b} = -2|\mathbf{a}||\mathbf{b}|\cos\theta$, so $\mathbf{a}\cdot\mathbf{b} = |\mathbf{a}||\mathbf{b}|\cos\theta$, and therefore $\cos\theta = \dfrac{\mathbf{a}\cdot\mathbf{b}}{|\mathbf{a}||\mathbf{b}|}$ — spec 7.7's formula, reached from the cosine rule rather than handed down as a rule to memorise. Two consequences follow immediately, and neither needs a separate justification once the identity above is derived. First, since $|\mathbf{a}|$ and $|\mathbf{b}|$ are magnitudes and therefore always positive for nonzero vectors, the SIGN of $\cos\theta$ is decided entirely by the sign of $\mathbf{a}\cdot\mathbf{b}$: a negative scalar product means an obtuse angle, full stop, whatever the individual components look like on the page. Second, if $\mathbf{a}\cdot\mathbf{b}=0$ with $\mathbf{a}$ and $\mathbf{b}$ both nonzero, then $\cos\theta = 0$ and $\theta = 90°$ — perpendicularity is not a separate fact to memorise, it is $\theta=90°$ read straight off the same identity, exactly matching spec 7.7's own guidance. One distinction matters before this gets used for real: $\theta = \angle AOB$ above sits at a SPECIFIC, real vertex, with $\mathbf{a}$ and $\mathbf{b}$ pointing along genuine edges of a genuine triangle, and it can come out obtuse and stay obtuse — there is nothing to "correct." A differently-phrased question — "find the angle between line $l_1$ and line $l_2$," with no named vertex at all — is a different situation: a LINE has no built-in direction (its direction vector $\mathbf{b}$ and $-\mathbf{b}$ describe exactly the same line), so the raw $\theta$ computed from whichever direction vectors happen to be written down is not itself a property of the two lines — only its acute value, in $[0°,90°]$, is. On a between-two-lines question, an obtuse raw result is converted by taking $180°-\theta$; on a named-vertex question like $\angle ABC$, the raw result IS the answer, obtuse or not. The two situations produce opposite-looking mistakes from the same confusion — "correcting" a genuinely obtuse vertex angle to acute, or leaving an uncorrected obtuse value on a between-two-lines question — so which rule applies depends entirely on which of the two the question actually asked.

Traps — 5

wrong-vectors-selected-for-the-angle-calculation

The most consistently-documented specific error anywhere in this topic, confirmed independently across three series. January 2021, Q2(a): "It was unclear whether candidates were finding angle BAC rather than the requested angle ABC or were just careless in not using BA.BC and used AC.BC... leading to an acute angle... Unfortunately, many candidates gave the acute angle 67.35° as their final answer." October 2022, Q3(b): "Not all candidates selected the correct directions for their vectors to give the obtuse angle, although some found the acute angle then subtracted from 180°." January 2024, Q6(c): "use of incorrect direction vectors was fairly common, often using position vectors especially that of the point of intersection found in (b)." Which specific vectors get picked up wrong varies — sometimes the wrong pair of direction vectors at the shared vertex, sometimes a position vector substituted for a direction vector entirely — but the underlying failure is identical every time: writing down SOME vectors that produce a plausible-looking number, without checking they are the two vectors the angle actually sits between.

acute-angle-reported-when-obtuse-required

Confirmed independently in two of the three series quoted above — not all three, and this lesson keeps that narrower count rather than rounding it up to match the trap above. January 2021, Q2(a) records candidates who, having used the wrong pair of vectors, "gave the acute angle 67.35° as their final answer" when the real geometry required an obtuse one. October 2022, Q3(b) records the same symptom directly: "Not all candidates selected the correct directions for their vectors to give the obtuse angle," while also noting a partial fix some candidates applied — "some found the acute angle then subtracted from 180°." A negative scalar product is not an inconvenience to argue away: it is the entire content of the answer, for a genuine named-vertex angle.

scalar-vs-vector-product-confusion

A conceptual error about what kind of object the scalar product even is, confirmed directly: "The most common error with the unsuccessful candidates was in their misunderstanding of a 'scalar' product and obtaining the vector (12,−10,24) rather than the value of 12−10+24" (Jan 2021, Q2(a)). The name is not decoration: $\mathbf{a}\cdot\mathbf{b} = a_1b_1+a_2b_2+a_3b_3$ is a SINGLE NUMBER, the sum of three products — never a vector with three separate components left sitting unadded.

skew-shown-via-not-perpendicular-instead-of-not-parallel

Confirmed directly on a real skew-lines question: "Many candidates did not fully understand the meaning of skew and only got as far as finding the values of μ and λ... the majority, however, failed to state or show that the given lines were not parallel. A few candidates, having said that the lines did not intersect, went on to show that they were not perpendicular" (Jan 2021, Q8). Skewness needs two conditions — not parallel, and not intersecting; perpendicularity answers neither of them, and substituting it in for the missing half earns nothing.

non-parallel-shown-via-non-identical-not-non-scalar-multiple

Confirmed directly on a real intersecting-or-skew question: "a fairly high proportion who achieved a pair of conflicting values failed to then conclude that this implied the lines did not intersect... Many candidates did not realise that they needed to consider the possibility of the lines being parallel. Of those that did, some concluded only that [direction vector 1] ≠ [direction vector 2] which was insufficient" (Oct 2022, Q9). Two direction vectors that are not written identically can still be scalar multiples of one another — and if they are, the lines ARE parallel regardless of how different the two vectors look on the page. The only valid test is the scalar-multiple check; a glance is not a check.

The card

a·b = a₁b₁+a₂b₂+a₃b₃ (NOT in the booklet — memorise). cosθ = a·b/(|a||b|). a·b=0 ⇔ a ⟂ b (a, b ≠ 0).
Line: r = a + tb (a = a point on the line, b = direction) or r = c + t(d − c), where d − c is the direction.
Parallel? Direction vectors must be scalar multiples — check EVERY component with the SAME k, not just some.
Not parallel: solve any 2 of the 3 component equations for the parameters, then check the 3rd. Consistent → intersecting (state the point). Inconsistent → skew.
Skew needs BOTH not-parallel AND not-intersecting shown — 'not perpendicular' proves neither.
Angle ABC at a named vertex: use BA and BC, report exactly what you get, obtuse or not. Angle BETWEEN two lines (no vertex): report the acute value, converting via 180°−θ if the raw result is obtuse.
Area of the parallelogram spanned by u, v from a shared vertex = |u||v|sinθ (triangle = half that); sinθ = √(1−cos²θ), reusing the same θ found from the scalar product — reversing either vector doesn't change the area, unlike the angle itself.
Cross product (NOT P4 spec content, but a real mark scheme credits it as a full-marks alternative): u×v = (u₂v₃−u₃v₂, u₃v₁−u₁v₃, u₁v₂−u₂v₁); |u×v| = |u||v|sinθ = the parallelogram's area directly, no angle needed.

Why it works — Where cos θ = a·b / (|a||b|) actually comes from — the cosine rule, not a new rule

Place the origin at O, so $\mathbf{a} = \overrightarrow{OA}$ and $\mathbf{b} = \overrightarrow{OB}$ are position vectors and $\theta = \angle AOB$. Triangle $OAB$ has sides $OA = |\mathbf{a}|$, $OB = |\mathbf{b}|$, and — by the 7.4 identity already used above — $AB = |\mathbf{b}-\mathbf{a}|$. The cosine rule, confirmed present in the formula booklet's P1 section (WMA14-verified-facts.md §3), states $AB^2 = OA^2 + OB^2 - 2\cdot OA\cdot OB\cos\theta$, i.e. $|\mathbf{b}-\mathbf{a}|^2 = |\mathbf{a}|^2+|\mathbf{b}|^2-2|\mathbf{a}||\mathbf{b}|\cos\theta$. Now expand the left side using the scalar product's own defining property, $|\mathbf{v}|^2 = \mathbf{v}\cdot\mathbf{v}$: $(\mathbf{b}-\mathbf{a})\cdot(\mathbf{b}-\mathbf{a}) = \mathbf{b}\cdot\mathbf{b} - 2\,\mathbf{a}\cdot\mathbf{b} + \mathbf{a}\cdot\mathbf{a} = |\mathbf{b}|^2 - 2\,\mathbf{a}\cdot\mathbf{b} + |\mathbf{a}|^2$. Setting the two expressions for $AB^2$ equal and cancelling the $|\mathbf{a}|^2+|\mathbf{b}|^2$ common to both sides leaves $-2\,\mathbf{a}\cdot\mathbf{b} = -2|\mathbf{a}||\mathbf{b}|\cos\theta$, so $\mathbf{a}\cdot\mathbf{b} = |\mathbf{a}||\mathbf{b}|\cos\theta$, and therefore $\cos\theta = \dfrac{\mathbf{a}\cdot\mathbf{b}}{|\mathbf{a}||\mathbf{b}|}$ — spec 7.7's formula, reached from the cosine rule rather than handed down as a rule to memorise. Two consequences follow immediately, and neither needs a separate justification once the identity above is derived. First, since $|\mathbf{a}|$ and $|\mathbf{b}|$ are magnitudes and therefore always positive for nonzero vectors, the SIGN of $\cos\theta$ is decided entirely by the sign of $\mathbf{a}\cdot\mathbf{b}$: a negative scalar product means an obtuse angle, full stop, whatever the individual components look like on the page. Second, if $\mathbf{a}\cdot\mathbf{b}=0$ with $\mathbf{a}$ and $\mathbf{b}$ both nonzero, then $\cos\theta = 0$ and $\theta = 90°$ — perpendicularity is not a separate fact to memorise, it is $\theta=90°$ read straight off the same identity, exactly matching spec 7.7's own guidance. One distinction matters before this gets used for real: $\theta = \angle AOB$ above sits at a SPECIFIC, real vertex, with $\mathbf{a}$ and $\mathbf{b}$ pointing along genuine edges of a genuine triangle, and it can come out obtuse and stay obtuse — there is nothing to "correct." A differently-phrased question — "find the angle between line $l_1$ and line $l_2$," with no named vertex at all — is a different situation: a LINE has no built-in direction (its direction vector $\mathbf{b}$ and $-\mathbf{b}$ describe exactly the same line), so the raw $\theta$ computed from whichever direction vectors happen to be written down is not itself a property of the two lines — only its acute value, in $[0°,90°]$, is. On a between-two-lines question, an obtuse raw result is converted by taking $180°-\theta$; on a named-vertex question like $\angle ABC$, the raw result IS the answer, obtuse or not. The two situations produce opposite-looking mistakes from the same confusion — "correcting" a genuinely obtuse vertex angle to acute, or leaving an uncorrected obtuse value on a between-two-lines question — so which rule applies depends entirely on which of the two the question actually asked.

Traps — 5

wrong-vectors-selected-for-the-angle-calculation

The most consistently-documented specific error anywhere in this topic, confirmed independently across three series. January 2021, Q2(a): "It was unclear whether candidates were finding angle BAC rather than the requested angle ABC or were just careless in not using BA.BC and used AC.BC... leading to an acute angle... Unfortunately, many candidates gave the acute angle 67.35° as their final answer." October 2022, Q3(b): "Not all candidates selected the correct directions for their vectors to give the obtuse angle, although some found the acute angle then subtracted from 180°." January 2024, Q6(c): "use of incorrect direction vectors was fairly common, often using position vectors especially that of the point of intersection found in (b)." Which specific vectors get picked up wrong varies — sometimes the wrong pair of direction vectors at the shared vertex, sometimes a position vector substituted for a direction vector entirely — but the underlying failure is identical every time: writing down SOME vectors that produce a plausible-looking number, without checking they are the two vectors the angle actually sits between.

acute-angle-reported-when-obtuse-required

Confirmed independently in two of the three series quoted above — not all three, and this lesson keeps that narrower count rather than rounding it up to match the trap above. January 2021, Q2(a) records candidates who, having used the wrong pair of vectors, "gave the acute angle 67.35° as their final answer" when the real geometry required an obtuse one. October 2022, Q3(b) records the same symptom directly: "Not all candidates selected the correct directions for their vectors to give the obtuse angle," while also noting a partial fix some candidates applied — "some found the acute angle then subtracted from 180°." A negative scalar product is not an inconvenience to argue away: it is the entire content of the answer, for a genuine named-vertex angle.

scalar-vs-vector-product-confusion

A conceptual error about what kind of object the scalar product even is, confirmed directly: "The most common error with the unsuccessful candidates was in their misunderstanding of a 'scalar' product and obtaining the vector (12,−10,24) rather than the value of 12−10+24" (Jan 2021, Q2(a)). The name is not decoration: $\mathbf{a}\cdot\mathbf{b} = a_1b_1+a_2b_2+a_3b_3$ is a SINGLE NUMBER, the sum of three products — never a vector with three separate components left sitting unadded.

skew-shown-via-not-perpendicular-instead-of-not-parallel

Confirmed directly on a real skew-lines question: "Many candidates did not fully understand the meaning of skew and only got as far as finding the values of μ and λ... the majority, however, failed to state or show that the given lines were not parallel. A few candidates, having said that the lines did not intersect, went on to show that they were not perpendicular" (Jan 2021, Q8). Skewness needs two conditions — not parallel, and not intersecting; perpendicularity answers neither of them, and substituting it in for the missing half earns nothing.

non-parallel-shown-via-non-identical-not-non-scalar-multiple

Confirmed directly on a real intersecting-or-skew question: "a fairly high proportion who achieved a pair of conflicting values failed to then conclude that this implied the lines did not intersect... Many candidates did not realise that they needed to consider the possibility of the lines being parallel. Of those that did, some concluded only that [direction vector 1] ≠ [direction vector 2] which was insufficient" (Oct 2022, Q9). Two direction vectors that are not written identically can still be scalar multiples of one another — and if they are, the lines ARE parallel regardless of how different the two vectors look on the page. The only valid test is the scalar-multiple check; a glance is not a check.

The card

V = π∫ y² dx, about the x-axis ONLY. π∫x²dy is not required on this paper.
Parametric: dx = (dx/dθ)dθ, so V = π∫ y² (dx/dθ) dθ — convert the θ-limits too, not just dx.
Squaring a compound y: (p±q)² = p² ± 2pq + q². Never drop the cross term.
y = sinθ ± cosθ type: y² = 1 ± sin2θ, from sin²θ+cos²θ=1 and 2sinθcosθ=sin2θ.
Bare sin²θ or cos²θ (not from squaring a compound sum): sin²θ = (1 − cos2θ)/2, cos²θ = (1 + cos2θ)/2 — makes an otherwise-unintegrable bare even power integrable.
Stuck on the integral: check for a reverse chain rule shape (f′(x)[f(x)]ⁿ) before reaching for parts.

Why it works — Where π∫y²dx actually comes from — thin disks, not a rule to memorise

Cut the solid into thin vertical slices, each of thickness $\delta x$, taken at some x-value between $a$ and $b$. Provided $\delta x$ is small, one slice is almost exactly a cylinder: its two flat faces are (almost) parallel circles, both of radius $y=f(x)$ — because every point on the boundary curve traces out a circle of that radius once it's rotated — and its length is $\delta x$. A cylinder's volume is $\pi r^2 h$, so this one slice has volume approximately $\pi y^2 \delta x$. Add up every slice from $x=a$ to $x=b$: $V \approx \sum \pi y^2 \delta x$. That sum is only an approximation, because $y$ changes slightly across each slice's own thickness — but the approximation gets better the thinner the slices are, and in the limit as $\delta x\to0$ the sum becomes exactly the definite integral $V=\pi\int_a^b y^2\,dx$. Nobody chose this formula for its own sake; it's what 'add up infinitely many infinitesimally thin cylinders' turns into once the adding is written as an integral instead of a sum — the same limiting argument that turns a sum of rectangle areas into $\int f(x)\,dx$ in the first place, just with a rectangle's area $f(x)\,\delta x$ replaced by a disk's volume $\pi y^2\,\delta x$. This is also the entire reason the formula squares $y$ rather than using it directly: a disk's area depends on the square of its radius, not the radius itself.

Traps — 5

wrong-constant-or-missing-pi

Confirmed directly on a real parametric volume-of-revolution question: "A few candidates mistakenly recalled the volume as 2π∫y²dx... as were those candidates who omitted the π" (Jan 2021, Q9). The mark scheme distinguishes the two: a wrong constant multiplier in front of an otherwise-correct integral is a different, more survivable error than losing the π-formula structure entirely — genuinely useful for partial-credit strategy, not just a warning to be careful. The fix is the same disk-method check the mechanism block above derives from scratch: the π comes from a single disk's own area, $\pi r^2$, once — not from the $2\pi$ of a full rotation, which is already accounted for by every point on the boundary sweeping out one full circle, not by an extra factor stacked on top.

double-angle-identity-not-applied-when-squaring-y

Confirmed on the same question: "There were a lot of slips in working with y² with many candidates failing to show that they had used sin2θ = 2sinθcosθ" (Jan 2021, Q9). Whenever $y(\theta)$ is a sum or difference of $\sin\theta$ and $\cos\theta$ terms, squaring it produces a cross term of the form $2\sin\theta\cos\theta$ — and that term cannot be integrated in that form. It has to be recognised as $\sin2\theta$ (or, depending on the setup, absorbed via $\sin^2\theta+\cos^2\theta=1$) before the integral is even attempted. This is a P3 identity, carried forward in the exam formula booklet under its own cumulative rule, not new P4 content; the P4-specific failure is not recognising WHEN it needs to be reached for. The SAME identity is also needed in the opposite direction on this same cited question: whenever $y(\theta)$ already contains a $\sin2\theta$ (or $\cos2\theta$) term in its own right and $\frac{dx}{d\theta}$ is a trig expression rather than a constant (e.g. $\sec^2\theta$, from $x=\tan\theta$), the double angle has to be EXPANDED — not condensed — to expose a $\cos^2\theta$ factor that then cancels against the $\sec^2\theta$; the worked chain further down models exactly this. On the real Jan 2021 Q9, expanding sin2θ this way is only the first of two separate dependent marks — the second is the power-reduction identity for a bare $\sin^2\theta$, a genuinely distinct technique named in its own trap-taxonomy item below, not a repeat of this one.

power-reduction-identity-not-used-for-a-bare-sin-or-cos-squared

A second, separate dM1 on the same cited question, earned independently of the double-angle expansion above: "Attempts to use sin²θ=(1−cos2θ)/2 [or cos²θ=(1+cos2θ)/2] ... and obtains Volume=∫(P±Qcos2θ)dθ. Depends on the first M." (Jan 2021, Q9). A BARE $\sin^2\theta$ or $\cos^2\theta$ — one that is not the leftover of squaring a compound sum, and so has no $\sin^2\theta+\cos^2\theta=1$ shortcut to reach for — has no elementary antiderivative in that form; power-reduction is what makes it integrable, rewriting it as a constant plus a $\cos2\theta$ term. It is easy to mistake this for "the double-angle step already done above" and skip it as redundant — it is not: the cross-term identity CONDENSES two terms into one, while this one converts a single even power into an integrable linear combination. Both marks appear in sequence on the same real question; each is earned or lost independently of the other.

wrong-integration-method-chosen-for-the-resulting-fraction

Confirmed on a different series' question, testing the same spec point via a Cartesian rather than parametric curve: "The major stumbling block for the majority of candidates was a failure to choose a correct approach to the integration which was at the heart of the question... few candidates made this choice and wasted much time pursuing incorrect methods which included... using integration by parts [and] integrating to ln(2x²+3)³" (Oct 2022, Q5). The examiner names the actually-efficient route explicitly: "recognise the integrand as being the result of a chain rule differentiation of (2x²+3)⁻² or else using a substitution." Both are shown, side by side, in the method comparison above — the report's own point is that a candidate reaching for integration by parts on this shape of integrand is reaching for a real, legitimate technique applied to the wrong problem, not making an arithmetic slip.

compound-expression-not-squared-correctly

Confirmed on a third, independent series: "Many wrote the correct formula... but made no further progress, usually due to being unable to see how to square y" (Jan 2024, Q7(b)). Skill 1 — writing $V=\pi\int y^2\,dx$ — was intact; the block was purely algebraic, squaring a $y$ that is itself a sum or difference of two terms rather than a single one. $(p+q)^2=p^2+2pq+q^2$: the cross term $2pq$ is not optional, and it is the term that most often goes missing, because dropping it still leaves something on the page that looks like a plausible squared expression. The marked solution above models exactly this failure and what it costs.

The card

(1+x)ⁿ = 1 + nx + n(n−1)/2! x² + n(n−1)(n−2)/3! x³ + … — booklet-supplied, valid ONLY for |x| < 1, n rational.
(a+bx)ⁿ = aⁿ(1 + (b/a)x)ⁿ. u = (b/a)x, sign included. Range: |u| < 1, i.e. |x| < |a/b|.
Scale EVERY term by the outer aⁿ — not just the first term of the bracket.
n(n−1)(n−2)… never hits zero unless n is a non-negative integer — that's why the series is infinite.
Approximating a value: substitute x into the ORIGINAL expression too, not just the series — both sides, not one.

Why it works — Why the series never terminates unless n is a non-negative integer

Every coefficient in the general expansion is built from the same running product: the coefficient of $x^r$ is $\frac{n(n-1)(n-2)\cdots(n-r+1)}{r!}$ — $r$ factors, counting down from $n$ in steps of 1. For $n$ a non-negative integer, this product reaches $n-n=0$ exactly once $r$ climbs to $n+1$ — one of the $r$ factors IS zero at that point, because the counting-down sequence started at a whole number and steps down in whole numbers, so it lands on zero precisely. Every term from $r=n+1$ onward is multiplied by that same zero and vanishes, which is exactly why the P2 expansion of $(1+x)^5$ has exactly 6 terms and then stops. For $n$ rational but NOT a non-negative integer — a fraction like $\frac12$, or a negative number like $-2$ — the counting-down sequence $n, n-1, n-2, \ldots$ never lands on exactly zero, because you'd need $n$ itself to be a whole number that the count-down reaches, and a fraction or a negative number reached by subtracting whole numbers from $n$ never equals zero. So no factor in the product is ever exactly zero, no term of the series is ever exactly zero, and the sum genuinely never stops. This is also exactly where the $|x|<1$ condition comes from: an infinite sum of terms only settles on one fixed value if the terms shrink fast enough as $r$ grows, and $|x|<1$ is precisely the condition that guarantees the $u^r$ factor inside each term shrinks toward zero rather than blowing up. Outside that range there is nothing algebraically wrong with $(1+x)^n$ itself — it's still a perfectly well-defined number — it just stops being equal to the sum of this particular infinite series.

Traps — 6

linear-coefficient-sign-lost

Confirmed directly: "Where errors did occur, they were usually the result of using 'x' or '4x' in the place of '−4x' in the expansion" (Jan 2024, Q1). Mechanically, this is a dropped sign when identifying $u$ for a bracket like $(1-4x)^n$ — and it is genuinely easy to miss on a self-check, because (as the sign-flip mechanism the second MCQ below walks through shows) dropping the sign on $u$ flips every ODD-power term of the expansion but leaves every EVEN-power term — including the constant, which is often the only term a rushed check re-reads — numerically identical to the correct version.

constant-not-scaled-across-every-term

Confirmed as a specific wrong value: "A common incorrect 'x' term used was 5x/4" (Jan 2021, Q1), on an expansion involving $\left(\frac14-5x\right)^{\frac12}$ after factoring. The most likely mechanism — the bank itself records only the symptom, not the cause, so this is offered as the probable explanation rather than a confirmed one — is a fraction-division slip independent of binomial content: dividing $-5x$ by $\frac14$ means MULTIPLYING by 4 (giving $-20x$), and the reversed operation, dividing by 4 instead, produces exactly $-\frac{5x}{4}$ — the precise wrong term the report names. The general lesson survives even if the exact mechanism is only probable: whatever constant gets factored out has to be correctly divided into (or multiplied through) the linear term too, not just applied to the constant term of the bracket.

x-squared-term-collapsed-into-x

Confirmed directly, alongside the harder-but-valid alternative some candidates reach for instead: "it was surprising to see how many candidates were confused by the 4x² term, with some replacing it with 4x whilst others attempting a more difficult (1−2x)^½ × (1+2x)^½" (Oct 2021, Q4) — the exact two responses the method-comparison block above is built to contrast directly. Worth stating plainly alongside this: a later report records the same confusion measurably improved — "the idea of expanding a binomial expansion in x² did not cause the same issues as last year, so clearly candidates have learned from previous series" (Oct 2022, Q4). That is a single, separate observation about a DIFFERENT series, not a second confirmation of the same error recurring — it is cited here as evidence that this specific confusion is fixable with direct practice, not as a second instance of the trap itself.

range-of-validity-not-derived-from-u

Confirmed as a specific set of wrong final answers: incorrect responses of "|x| < 4, x < 2 and |x| < ±2" (Oct 2022, Q4) "when the correct condition should have been derived from the |ax|<1 form." Each of the three wrong patterns is a different way of skipping the actual derivation — using a coefficient directly as the bound instead of its reciprocal, dropping the modulus bars and keeping only one side of the inequality, or writing a ± next to a modulus that already covers both signs and describes nothing extra by doing so. All three are avoided the same way: write $|u|<1$ first, in terms of whatever $u$ actually is for THIS question, and solve that inequality — never read a bound off the original coefficients by pattern-matching a remembered shape.

substituted-into-only-one-side-of-the-approximation

Confirmed directly, and among the most severe outcomes recorded anywhere in the whole research bank for this qualification: "scores of 0 marks were very common with x = ¼ being substituted into only one side of the expansion. It was important to see x = ¼ being substituted into both sides of the expansion" (Oct 2021, Q4). The mechanism: an approximation question is built on an identity — the original unexpanded expression equals the series, at any x inside the range of validity — and substituting x into only the series produces a bare number, disconnected from whatever surd or value it was supposed to approximate. Substituting the same x into the ORIGINAL expression too is what establishes what the number computed from the series is actually an approximation OF; skipping it is treated as though the question was never actually answered, not just answered imprecisely.

decimal-given-when-exact-form-required

Confirmed directly: "A very small minority of candidates unfortunately gave a decimal approximation for their answer" (Jan 2021, Q1) on a question requiring an exact surd or fraction. This connects to a general Pure Mathematics marking principle documented across the whole qualification, not specific to this spec point: where an exact answer is asked for, marks are normally lost for resorting to a rounded decimal instead. When a question's final instruction asks for an exact form — a fraction, a surd, a value 'in the form p/q' — the exact form is the deliverable; a decimal that rounds to the same number is a different, lower-credit answer, not an equivalent way of writing the same thing.

The card

Distinct linear factors (ax+b)(cx+d)(ex+f): one term each, found by cover-up (substitute each factor's own root).
Repeated factor (cx+d)²: TWO terms — B/(cx+d) + C/(cx+d)² — never one. A denominator of degree n needs n unknowns total.
Numerator degree ≥ denominator degree: divide first, or use the identity method with an added constant — A/(x-a)+B/(x-b) alone can't represent it.
∫1/(ax+b)dx = (1/a)ln|ax+b| + c. The 1/a is real and load-bearing — confirmed the single most-documented slip in this topic, in TWO series.
∫1/(ax+b)ⁿdx for n≥2 is POWER-form, no logarithm at all: -1/[(n-1)a(ax+b)ⁿ⁻¹] + c.
x/(x²+5)-style integrands are NOT partial fractions (quadratic factors excluded by spec 2.1) — they integrate by recognition instead.
Expanding a decomposition in x (spec 4.1): factor a constant out of EVERY term to reach k(1+u)ⁿ, same as for a single bracket. The combined range of validity is the SMALLEST of the individual |u|<1 ranges, not the largest.

Why it works — Why a partial-fraction identity can be checked at chosen values of x — and why matching coefficients works too

A partial-fraction decomposition is written with the symbol ≡, not =, and that distinction is the entire justification for both standard solving methods. An equation like $2x=6$ is only true for one value of $x$; an IDENTITY like $8x^2+8x-3\equiv A(x+1)(2x+1)+B(x-2)(2x+1)+C(x-2)(x+1)$ is claimed to be true for every single value of $x$ at once — which means it is safe to substitute ANY value into it, including a value that makes one of the original fraction's own denominators zero (the identity itself, once cleared of denominators, is an ordinary polynomial equation defined everywhere, even where the original fraction wasn't). Substituting the root of one bracket makes every OTHER bracket's term vanish on the right-hand side, leaving an equation in only one unknown — this is the cover-up rule, and it works precisely because the identity is guaranteed to hold at that value too, not despite it. A second, completely independent consequence of the same ≡ symbol: two polynomials that are equal for every value of $x$ must have identical coefficients of every power of $x$ — $x^2$, $x^1$, $x^0$ separately — since a polynomial is entirely determined by its coefficients. This is why comparing coefficients is an equally valid, equally rigorous route to the same unknowns, not an approximation or a shortcut: both methods are reading off different, but equally true, consequences of one ≡ statement.

Traps — 6

extra-power-created-by-over-multiplying-a-repeated-factor

A real examiner report on a repeated-linear-factor decomposition (a DIFFERENT question from this lesson's own (3x+1)(x−2)² example, with its own denominator built around a bracket of (2x+1)) records: "Relatively few students made the error of multiplying by the product of the three denominators so having (2x+1)³" (Jan 2024, Q2(a)) — i.e. treating the repeated bracket as though it needed multiplying by itself an extra, unneeded time when clearing denominators, turning a correct squared bracket into a wrong cubed one. Real, but a genuine minority error ("relatively few"), not the dominant failure mode on this question type — that distinction is worth keeping honest rather than inflating.

simultaneous-equations-more-error-prone-than-substitution

A real examiner report on the same question records a genuine, quotable method-choice recommendation: simultaneous equations built from expanded coefficients, used by a significant number of candidates, were "more prone to error than attempts via substitution" (Jan 2024, Q2(a)) — exactly the finding this lesson's own method-comparison block anchors on. Both methods are legitimate and both reach full marks; the real record's point is which one real scripts actually get wrong more often, not that one method is invalid.

sign-lost-in-a-long-division-remainder

On a question comparing the identity method against long division as two legitimate routes to the same decomposition: "The most common error with candidates who used this method was to set the numerator of the partial fraction equal to 6 rather than -6 and then forgetting the negative when the values were put back into the expression" (Oct 2021, Q3(a)) — a sign lost while forming the remainder during division, then lost a SECOND time when substituting values back in. This lesson's own marked-solution above recreates the same failure class with its own numbers (2x+7 corrupted to 2x−1), not the real question's own 6/−6.

1-over-a-scaling-factor-omitted-on-a-ln-integral

The single most robustly documented error in this whole topic — confirmed independently in TWO separate series: "a common error was to omit the ½ in ∫1/(2x−1) dx = ½ ln(2x−1)" (Oct 2022, Q2), and again, on a different question testing the same skill: "A number of students failed to divide the second logarithm by two" (Jan 2024, Q2(b)). Both quotes describe the identical omission — the reciprocal of the x-coefficient inside the bracket, dropped from the front of the logarithm — this lesson's own marked-solution above is built specifically to anchor on this two-series-confirmed finding.

ln-rule-misapplied-to-a-repeated-factor

Flagged honestly as VERIDIAN-original pedagogical inference, NOT a sourced examiner-report finding — no quote in this unit's research bank documents this specific confusion. The two integrand shapes $\frac{1}{ax+b}$ and $\frac{1}{(ax+b)^2}$ look almost identical on the page, and it would be a natural mistake to apply the ln rule to BOTH, writing something like $\ln|(x-2)^2|$ for the squared case instead of the correct power-rule answer $\frac{-1}{x-2}$. Included here because the two integrand shapes really are this visually close, and this lesson's own mechanism block above exists specifically to derive why they need genuinely different rules — not because a real report has been found documenting students making exactly this error on this paper.

degree-condition-overlooked-before-attempting-the-basic-form

Anchored to spec 2.1's own explicit clause — "The degree of the numerator may equal or exceed the degree of the denominator" — rather than to a specific examiner-report quote describing students missing this: no citation in this unit's research bank documents this exact failure in these terms. Attempting $\frac{A}{x-a}+\frac{B}{x-b}$ directly on a fraction whose numerator's degree already matches the denominator's leads nowhere, because no choice of A and B can make that form reproduce a fraction that doesn't tend to 0 as x grows large. Worth checking degree BEFORE choosing a decomposition form, not after the first attempt fails.

The card

∫u(dv/dx)dx = uv − ∫v(du/dx)dx. Given in the booklet — the job is choosing u and dv/dx, not recalling the formula.
u gets simpler when differentiated (polynomials, ln x). dv/dx must be integrable and stay 'the same kind' (eˣ, sin, cos).
∫ln x dx: treat as (ln x)×1, u=ln x, dv/dx=1 → x ln x − x + c.
Repeated parts: track every constant through every term. Cyclic case (eˣsinx-type): differentiate the SAME factor both times, then solve the resulting equation for the integral.
Substitution replaces dx with (dx/du)du — never just relabels it. On a definite integral, convert the LIMITS too, via the same substitution applied to the original x-limits.

Why it works — Where the integration by parts formula actually comes from

Start from the product rule, exactly as stated above: $\frac{d}{dx}[uv] = u\frac{dv}{dx} + v\frac{du}{dx}$. Integrate both sides with respect to $x$. The left-hand side integrates back to exactly what was differentiated — integration undoes differentiation, so $\int \frac{d}{dx}[uv]\,dx = uv$. The right-hand side splits, since integration distributes over a sum: $\int u\frac{dv}{dx}\,dx + \int v\frac{du}{dx}\,dx$. So $uv = \int u\frac{dv}{dx}\,dx + \int v\frac{du}{dx}\,dx$, and rearranging for the first integral gives $\int u\frac{dv}{dx}\,dx = uv - \int v\frac{du}{dx}\,dx$ — exactly the formula printed in the Mathematical Formulae and Statistical Tables booklet, confirmed verbatim against the booklet's own P4 section. This is precisely what the specification means by describing the two techniques in this lesson as "the reverse processes of the chain and product rules respectively": nothing here is a new rule invented for integration, it is the product rule, read backwards. Because the formula is printed, there is nothing in it to memorise — the entire content of the skill lives in the two choices the formula itself is silent on: which factor becomes $u$, and which becomes $\frac{dv}{dx}$.

Traps — 6

ln-x-squared-treated-as-interchangeable-with-2-ln-x-mid-derivation

Confirmed on a real repeated-parts question that reused an earlier part's result: "many candidates assumed that ln(x²) was identical to 2 ln x and failed to score" (Oct 2021, Q8(b)). The identity itself is true — ln(x²) ≡ 2ln|x| — so the failure is not the algebra, it is using it as an unannounced shortcut in the middle of a derivation the mark scheme is following line by line; a substitution the examiner cannot see the justification for breaks the chain of reasoning being credited, even when the number it produces is correct. Write out any such rewrite as its own explicit line, not an invisible mental step.

coefficient-and-differentiation-slips-in-a-second-parts-application

A real second application of parts, building on an earlier part of the same question, records two separate documented failures at once: "A common error was then to integrate/differentiate the term cos2x incorrectly, losing the accuracy mark," and, further into the same question, "Some candidates failed to realise they required a factor of 3 within the final integral" (both Oct 2022, Q7(ii)). Together they describe exactly the risk the traditional method in the method comparison above is built to make visible: a SECOND application of the formula has to re-differentiate or re-integrate correctly all over again, and any constant carried from the first application has to survive being distributed across every term the second application produces — not just the first one it touches.

inverse-function-misapplied-to-a-substitution-term

Confirmed on a real substitution question: examiners record "candidates who changed sin 2x to sin⁻¹((u−3)/4)" (Oct 2021, Q6) — applying an inverse trig function to part of the substitution instead of substituting directly for the variable. A substitution replaces x, and every function of x, with an expression in u by direct algebraic rearrangement of the substitution itself; nothing about the technique ever calls for an inverse function, and reaching for one is a sign the substitution's own rearrangement — solving it for x, not for some other quantity — was skipped.

dx-relabelled-as-du-without-being-converted

Confirmed on the same question: "candidates who ignored the dx and simply wrote it as du" (Oct 2021, Q6) — one of the most fundamental substitution failures in this facts bank, and the exact failure the ∫x√(x−2)dx worked chain above is built to demonstrate the opposite of. dx is never simply crossed out and replaced with du; it is replaced by the FULL expression dx/du × du, found by differentiating the substitution. Dropping that factor is not a rounding error — it deletes the entire chain-rule content of the technique, leaving an expression that only coincidentally resembles the correct one when the factor happens to equal 1.

limits-of-integration-not-converted-to-the-new-variable

Confirmed on a real definite-integral substitution question: examiners record marks lost to "use of the x limits, ln7 and ln5 instead of the u limits of 4 and 2" (Oct 2022, Q7(i)) — evaluating the u-form antiderivative at the original x-values instead of converting them first. The marked solution above is built around exactly this failure: once the variable of integration changes, the two numbers at the top and bottom of the integral sign describe values of the new variable or the old one, never a mix of both.

a-numerical-factor-lost-within-the-substituted-expression

Confirmed on the same question, as a separate failure from the limits error above: "losing the factor 4 in the expression" (Oct 2022, Q7(i)). This is the coefficient-tracking risk from earlier in this lesson, arriving from the substitution side rather than the repeated-parts side: whenever dx/du is anything other than 1, that factor has to be carried through every remaining line of working, not dropped the moment it stops being the newest thing written down.

The card

∫u(dv/dx)dx = uv − ∫v(du/dx)dx. Given in the booklet — the job is choosing u and dv/dx, not recalling the formula.
u gets simpler when differentiated (polynomials, ln x). dv/dx must be integrable and stay 'the same kind' (eˣ, sin, cos).
∫ln x dx: treat as (ln x)×1, u=ln x, dv/dx=1 → x ln x − x + c.
Repeated parts: track every constant through every term. Cyclic case (eˣsinx-type): differentiate the SAME factor both times, then solve the resulting equation for the integral.
Substitution replaces dx with (dx/du)du — never just relabels it. On a definite integral, convert the LIMITS too, via the same substitution applied to the original x-limits.

Why it works — Where the integration by parts formula actually comes from

Start from the product rule, exactly as stated above: $\frac{d}{dx}[uv] = u\frac{dv}{dx} + v\frac{du}{dx}$. Integrate both sides with respect to $x$. The left-hand side integrates back to exactly what was differentiated — integration undoes differentiation, so $\int \frac{d}{dx}[uv]\,dx = uv$. The right-hand side splits, since integration distributes over a sum: $\int u\frac{dv}{dx}\,dx + \int v\frac{du}{dx}\,dx$. So $uv = \int u\frac{dv}{dx}\,dx + \int v\frac{du}{dx}\,dx$, and rearranging for the first integral gives $\int u\frac{dv}{dx}\,dx = uv - \int v\frac{du}{dx}\,dx$ — exactly the formula printed in the Mathematical Formulae and Statistical Tables booklet, confirmed verbatim against the booklet's own P4 section. This is precisely what the specification means by describing the two techniques in this lesson as "the reverse processes of the chain and product rules respectively": nothing here is a new rule invented for integration, it is the product rule, read backwards. Because the formula is printed, there is nothing in it to memorise — the entire content of the skill lives in the two choices the formula itself is silent on: which factor becomes $u$, and which becomes $\frac{dv}{dx}$.

Traps — 6

ln-x-squared-treated-as-interchangeable-with-2-ln-x-mid-derivation

Confirmed on a real repeated-parts question that reused an earlier part's result: "many candidates assumed that ln(x²) was identical to 2 ln x and failed to score" (Oct 2021, Q8(b)). The identity itself is true — ln(x²) ≡ 2ln|x| — so the failure is not the algebra, it is using it as an unannounced shortcut in the middle of a derivation the mark scheme is following line by line; a substitution the examiner cannot see the justification for breaks the chain of reasoning being credited, even when the number it produces is correct. Write out any such rewrite as its own explicit line, not an invisible mental step.

coefficient-and-differentiation-slips-in-a-second-parts-application

A real second application of parts, building on an earlier part of the same question, records two separate documented failures at once: "A common error was then to integrate/differentiate the term cos2x incorrectly, losing the accuracy mark," and, further into the same question, "Some candidates failed to realise they required a factor of 3 within the final integral" (both Oct 2022, Q7(ii)). Together they describe exactly the risk the traditional method in the method comparison above is built to make visible: a SECOND application of the formula has to re-differentiate or re-integrate correctly all over again, and any constant carried from the first application has to survive being distributed across every term the second application produces — not just the first one it touches.

inverse-function-misapplied-to-a-substitution-term

Confirmed on a real substitution question: examiners record "candidates who changed sin 2x to sin⁻¹((u−3)/4)" (Oct 2021, Q6) — applying an inverse trig function to part of the substitution instead of substituting directly for the variable. A substitution replaces x, and every function of x, with an expression in u by direct algebraic rearrangement of the substitution itself; nothing about the technique ever calls for an inverse function, and reaching for one is a sign the substitution's own rearrangement — solving it for x, not for some other quantity — was skipped.

dx-relabelled-as-du-without-being-converted

Confirmed on the same question: "candidates who ignored the dx and simply wrote it as du" (Oct 2021, Q6) — one of the most fundamental substitution failures in this facts bank, and the exact failure the ∫x√(x−2)dx worked chain above is built to demonstrate the opposite of. dx is never simply crossed out and replaced with du; it is replaced by the FULL expression dx/du × du, found by differentiating the substitution. Dropping that factor is not a rounding error — it deletes the entire chain-rule content of the technique, leaving an expression that only coincidentally resembles the correct one when the factor happens to equal 1.

limits-of-integration-not-converted-to-the-new-variable

Confirmed on a real definite-integral substitution question: examiners record marks lost to "use of the x limits, ln7 and ln5 instead of the u limits of 4 and 2" (Oct 2022, Q7(i)) — evaluating the u-form antiderivative at the original x-values instead of converting them first. The marked solution above is built around exactly this failure: once the variable of integration changes, the two numbers at the top and bottom of the integral sign describe values of the new variable or the old one, never a mix of both.

a-numerical-factor-lost-within-the-substituted-expression

Confirmed on the same question, as a separate failure from the limits error above: "losing the factor 4 in the expression" (Oct 2022, Q7(i)). This is the coefficient-tracking risk from earlier in this lesson, arriving from the substitution side rather than the repeated-parts side: whenever dx/du is anything other than 1, that factor has to be carried through every remaining line of working, not dropped the moment it stops being the newest thing written down.

The card

dy/dx = f(x)g(y) → separate: dy/g(y) = f(x)dx, then integrate — one constant, not two.
General solution has +C; a given (x,y) boundary condition turns it into the particular solution.
Connected rates: relate the two quantities first, differentiate that relation, then chain rule: dV/dt = dV/dh × dh/dt.
Booklet gives sphere surface area and cone curved surface area only — every volume formula is memory, not lookup.
Ask: is the geometric quantity actually constant, or is it itself a function of the variable you're differentiating?
State the unit the question was worked in from the start — a number alone is not a complete AO5 answer.

Why it works — Separating variables — why moving dy and dx around is legal, and why only one constant survives

\frac{dy}{dx} = f(x)g(y), rearranged and integrated as $\int \frac{1}{g(y)}\,dy = \int f(x)\,dx$, looks like it treats dy and dx as ordinary numbers you can shuffle across an equals sign — which they are not. Here is why the shuffle is legal anyway. Let $G(y)$ be any antiderivative of $\frac{1}{g(y)}$, so that $G'(y) = \frac{1}{g(y)}$ by definition. Now differentiate $G(y)$ with respect to x using the chain rule: $\frac{d}{dx}\left[G(y)\right] = G'(y)\cdot\frac{dy}{dx} = \frac{1}{g(y)}\cdot\frac{dy}{dx}$. But the original equation says $\frac{dy}{dx} = f(x)g(y)$, so substituting that in: $\frac{d}{dx}\left[G(y)\right] = \frac{1}{g(y)}\cdot f(x)g(y) = f(x)$ — the $g(y)$'s cancel completely, leaving a plain function of x on both sides of a genuine, ordinary equation: $\frac{d}{dx}\left[G(y)\right] = f(x)$. Integrating both sides with respect to x, in the completely standard sense (no shuffling required), gives $G(y) = \int f(x)\,dx + C$. And by G's own definition, $G(y)$ IS $\int \frac{1}{g(y)}\,dy$. So $\int \frac{1}{g(y)}\,dy = \int f(x)\,dx + C$ — exactly the "separate and integrate each side" recipe, but now derived rather than pattern-matched, and it also explains the one-constant rule directly: the constant only ever entered on ONE side of this derivation, the x-side, because the y-side is just $G(y)$ with nothing left to integrate independently. Writing a separate $+C_1$ on the y-side and a separate $+C_2$ on the x-side and then combining them into one C at the end gives the identical final equation — the two-constant version is not wrong, it is simply doing one extra step of bookkeeping to reach the same place.

Traps — 7

sphere-formula-used-for-a-cylinder

Confirmed directly on a real cylindrical rates-of-change question: candidates "using the volume of the cylinder as V = (4/3)πr²h" (Oct 2021, Q9) — the volume of a SPHERE, applied to a cylinder, and even then written with an h that the sphere formula does not have. Neither formula is in the exam formula booklet: WMA14-verified-facts.md §3 confirms the P1 mensuration entry gives only a sphere's surface area and a cone's curved surface area, never a volume for any shape. Every solid-volume formula in this topic has to be recognised correctly from memory, from the shape actually described.

two-term-rate-collapsed-to-one

Confirmed on the same question: candidates "using either dV/dt = 0.6π or dV/dt = −0.15πh" (Oct 2021, Q9) — each is exactly half of what should have been a single two-term rate (an inflow AND an outflow), with the other term silently dropped. The marked-solution above is built specifically so this error is checkable: dropping the inflow gives dh/dt = −h/50, which visibly fails to match the printed "show that" target.

constant-integrated-as-a-logarithm

Confirmed on the same question, at the solving stage: candidates "integrating 1/320 to ln(320t)" (Oct 2021, Q9) — a plain constant integrated with respect to time should give (constant)×t, not a logarithm of time. The analogous line in the marked-solution above is $\int \frac{1}{50}\,dt = \frac{t}{50}$; the documented wrong version would write $\ln(50t)$ instead.

answer-given-in-the-wrong-time-unit

Confirmed on the same question, at the very last step: candidates "giving the units for the answer 208 as seconds rather than minutes" (Oct 2021, Q9) — the number itself was right, and the mark was lost purely because the stated unit did not match the unit the whole question had been working in from its first line. This is exactly why the marked-solution's part (c) states "minutes" explicitly rather than leaving the reader to assume it.

denominator-mishandled-when-separating

Confirmed on a real separable-equation question: "many candidates who decided to 'move' the 4, ended up with an incorrect starting equation of ∫4/y² dy = ∫1/(4x+5)^(3/2) dx at some point" (Oct 2021, Q2) — a genuine rearrangement error before any integration has even begun, not a calculus mistake. The same report also documents a matching error at the OTHER end of the same question, after both sides had already been correctly integrated and solved for y: converting an equation of the form a/y = b√(4x+5) + c into y = 1/(a·b√(4x+5)) + 1/c — distributing a reciprocal across a sum, which is not a legal algebraic move (1/(p+q) is not 1/p + 1/q). One question, two independently documented ways to lose marks on pure algebra either side of a perfectly good middle section.

fudged-reverse-fit-on-a-show-that-answer

Confirmed on a real "show that a particular solution equals a printed expression" question: "the given answer here persuaded many candidates to 'adjust' their working following obvious mistakes" (Jan 2021, Q10). Because the target is printed on the page in an ag question, a derivation that quietly changes a wrong intermediate number to make the last line match is a real, examiner-documented pattern — not a hypothetical one — and it is specifically NOT credited even when the final line is correct, because the mark is for the derivation, not for the coincidence of matching text. The warrantCheck attached to the marked-solution's part (a) above is built directly around this trap.

related-quantity-treated-as-fixed-instead-of-substituted

Confirmed on a real cone-based rates-of-change question — surface area rather than volume, but the same connected-rates setup skill spec 5.2 covers: candidates "treating l in the given S formula as a constant" (Jan 2024, Q4), where l (the cone's slant height) was actually a function of the radius via Pythagoras and needed to be substituted as such BEFORE differentiating, not held fixed. The same report records the question as challenging for "a very large majority of students," with "the concept of rates of change... shown to be not well understood by most." The cylinder in this lesson's own worked example is the case where a quantity genuinely IS a constant (the cross-sectional radius, with vertical tank walls) — this trap is what happens when that assumption is applied to a shape where it no longer holds.

The card

dy/dx = f(x)g(y) → separate: dy/g(y) = f(x)dx, then integrate — one constant, not two.
General solution has +C; a given (x,y) boundary condition turns it into the particular solution.
Connected rates: relate the two quantities first, differentiate that relation, then chain rule: dV/dt = dV/dh × dh/dt.
Booklet gives sphere surface area and cone curved surface area only — every volume formula is memory, not lookup.
Ask: is the geometric quantity actually constant, or is it itself a function of the variable you're differentiating?
State the unit the question was worked in from the start — a number alone is not a complete AO5 answer.

Why it works — Separating variables — why moving dy and dx around is legal, and why only one constant survives

\frac{dy}{dx} = f(x)g(y), rearranged and integrated as $\int \frac{1}{g(y)}\,dy = \int f(x)\,dx$, looks like it treats dy and dx as ordinary numbers you can shuffle across an equals sign — which they are not. Here is why the shuffle is legal anyway. Let $G(y)$ be any antiderivative of $\frac{1}{g(y)}$, so that $G'(y) = \frac{1}{g(y)}$ by definition. Now differentiate $G(y)$ with respect to x using the chain rule: $\frac{d}{dx}\left[G(y)\right] = G'(y)\cdot\frac{dy}{dx} = \frac{1}{g(y)}\cdot\frac{dy}{dx}$. But the original equation says $\frac{dy}{dx} = f(x)g(y)$, so substituting that in: $\frac{d}{dx}\left[G(y)\right] = \frac{1}{g(y)}\cdot f(x)g(y) = f(x)$ — the $g(y)$'s cancel completely, leaving a plain function of x on both sides of a genuine, ordinary equation: $\frac{d}{dx}\left[G(y)\right] = f(x)$. Integrating both sides with respect to x, in the completely standard sense (no shuffling required), gives $G(y) = \int f(x)\,dx + C$. And by G's own definition, $G(y)$ IS $\int \frac{1}{g(y)}\,dy$. So $\int \frac{1}{g(y)}\,dy = \int f(x)\,dx + C$ — exactly the "separate and integrate each side" recipe, but now derived rather than pattern-matched, and it also explains the one-constant rule directly: the constant only ever entered on ONE side of this derivation, the x-side, because the y-side is just $G(y)$ with nothing left to integrate independently. Writing a separate $+C_1$ on the y-side and a separate $+C_2$ on the x-side and then combining them into one C at the end gives the identical final equation — the two-constant version is not wrong, it is simply doing one extra step of bookkeeping to reach the same place.

Traps — 7

sphere-formula-used-for-a-cylinder

Confirmed directly on a real cylindrical rates-of-change question: candidates "using the volume of the cylinder as V = (4/3)πr²h" (Oct 2021, Q9) — the volume of a SPHERE, applied to a cylinder, and even then written with an h that the sphere formula does not have. Neither formula is in the exam formula booklet: WMA14-verified-facts.md §3 confirms the P1 mensuration entry gives only a sphere's surface area and a cone's curved surface area, never a volume for any shape. Every solid-volume formula in this topic has to be recognised correctly from memory, from the shape actually described.

two-term-rate-collapsed-to-one

Confirmed on the same question: candidates "using either dV/dt = 0.6π or dV/dt = −0.15πh" (Oct 2021, Q9) — each is exactly half of what should have been a single two-term rate (an inflow AND an outflow), with the other term silently dropped. The marked-solution above is built specifically so this error is checkable: dropping the inflow gives dh/dt = −h/50, which visibly fails to match the printed "show that" target.

constant-integrated-as-a-logarithm

Confirmed on the same question, at the solving stage: candidates "integrating 1/320 to ln(320t)" (Oct 2021, Q9) — a plain constant integrated with respect to time should give (constant)×t, not a logarithm of time. The analogous line in the marked-solution above is $\int \frac{1}{50}\,dt = \frac{t}{50}$; the documented wrong version would write $\ln(50t)$ instead.

answer-given-in-the-wrong-time-unit

Confirmed on the same question, at the very last step: candidates "giving the units for the answer 208 as seconds rather than minutes" (Oct 2021, Q9) — the number itself was right, and the mark was lost purely because the stated unit did not match the unit the whole question had been working in from its first line. This is exactly why the marked-solution's part (c) states "minutes" explicitly rather than leaving the reader to assume it.

denominator-mishandled-when-separating

Confirmed on a real separable-equation question: "many candidates who decided to 'move' the 4, ended up with an incorrect starting equation of ∫4/y² dy = ∫1/(4x+5)^(3/2) dx at some point" (Oct 2021, Q2) — a genuine rearrangement error before any integration has even begun, not a calculus mistake. The same report also documents a matching error at the OTHER end of the same question, after both sides had already been correctly integrated and solved for y: converting an equation of the form a/y = b√(4x+5) + c into y = 1/(a·b√(4x+5)) + 1/c — distributing a reciprocal across a sum, which is not a legal algebraic move (1/(p+q) is not 1/p + 1/q). One question, two independently documented ways to lose marks on pure algebra either side of a perfectly good middle section.

fudged-reverse-fit-on-a-show-that-answer

Confirmed on a real "show that a particular solution equals a printed expression" question: "the given answer here persuaded many candidates to 'adjust' their working following obvious mistakes" (Jan 2021, Q10). Because the target is printed on the page in an ag question, a derivation that quietly changes a wrong intermediate number to make the last line match is a real, examiner-documented pattern — not a hypothetical one — and it is specifically NOT credited even when the final line is correct, because the mark is for the derivation, not for the coincidence of matching text. The warrantCheck attached to the marked-solution's part (a) above is built directly around this trap.

related-quantity-treated-as-fixed-instead-of-substituted

Confirmed on a real cone-based rates-of-change question — surface area rather than volume, but the same connected-rates setup skill spec 5.2 covers: candidates "treating l in the given S formula as a constant" (Jan 2024, Q4), where l (the cone's slant height) was actually a function of the radius via Pythagoras and needed to be substituted as such BEFORE differentiating, not held fixed. The same report records the question as challenging for "a very large majority of students," with "the concept of rates of change... shown to be not well understood by most." The cylinder in this lesson's own worked example is the case where a quantity genuinely IS a constant (the cross-sectional radius, with vertical tank walls) — this trap is what happens when that assumption is applied to a shape where it no longer holds.

The card

Eliminate t: solve ONE equation for t (or for cos t/sin t), substitute into the OTHER — or, if a target general form is given, substitute x(t) into it and derive every coefficient by matching it to y(t) as a genuine identity in t. Don't just guess coefficients to fit the target without deriving them.
Trig pair: isolate cos t and sin t FULLY (subtract, then divide) — THEN square both and add, using sin²t + cos²t ≡ 1.
A parametric elimination question is pure algebra — reaching for differentiation on it scores nothing (Oct 2022, Q1).
Domain (of x) comes from mapping the parameter's own bounds through x(t). Range (of y) is a separate, y-shaped question — never answer one with the other.
Closed, bounded t-domain (e.g. 1 ≤ t ≤ 5): map the domain by substituting the two literal endpoint values directly.
Open and/or unbounded t-domain (e.g. t > 0): there is no literal endpoint to substitute — reason about what x approaches as a LIMIT at each excluded or infinite end instead, then combine into an open/unbounded x-domain.
To find the range: check the two endpoint t-values AND look for an interior turning point (complete the square, or differentiate) — endpoints alone can miss an interior min/max.
Before using an interior turning point, check its t-value actually lies inside the stated domain — one outside it belongs to a part of the curve that was never drawn.
If g(x) is monotonic (no turning point exists at all — check g'(x) never changes sign), stop looking for one. Find the range's two boundaries instead: evaluate g at its own excluded endpoint directly (valid if g is continuous there), and find the limiting value as x → ∞ (for a rational g(x) = (ax+b)/(cx+d), this is a/c).

Why it works — Why the parameter's domain becomes the cartesian function's domain — and why its range needs more than two numbers

When $t$ is restricted to a domain $D$ — a CLOSED, bounded interval, say $a \leq t \leq b$, is the simplest case, covered first below, but $D$ can equally be OPEN and/or UNBOUNDED (such as $t>0$, which both excludes its own lower end and has no upper one), a genuinely different case covered later in this same block — the SET of $x$-values genuinely reached — $\{f(t) : t \in D\}$ — is exactly what defines the valid domain of the cartesian equation. The bare equation, as pure algebra with no reference to $t$, might make sense for a wider range of $x$; only the $x$-values a real point of the curve actually reaches belong to it. When $f$ is strictly monotonic over $D$ (a linear $x=f(t)$ with nonzero coefficient always is), the two endpoints of $D$ map straight onto the two endpoints of the $x$-domain, with every value in between covered too — so for a monotonic $f$, checking the two endpoints of $t$ genuinely is enough to state the whole $x$-domain, correctly. The range of $y$ is a structurally different question, and this is exactly where it stops being safe to apply the same shortcut. $y=g(t)$ is very often NOT monotonic over $D$ — a quadratic in $t$ has exactly one turning point — and if that turning point sits at some $t^\ast$ INTERIOR to $D$, the values $g$ takes near $t^\ast$ are not bounded between $g(a)$ and $g(b)$ at all: they can dip below both, or rise above both, at that one interior point alone. So the very shortcut that correctly finds the $x$-domain (because $x$ happens to be monotonic here) fails outright for the $y$-range whenever $y$ is not — precisely the failure a real examiner report records directly: "Common incorrect solutions followed attempts to substitute either end of the domain in the parametric equation for y. This usually resulted in only one of the two marks being scored, with the minimum value missing" (Oct 2021, Q5(c)). The fix reuses exactly the completing-the-square reading this lesson's own prerequisite lesson already establishes for a directly-given quadratic: rewrite $y$ as a function of $x$, find any interior turning point (by completing the square, or by setting $\frac{dy}{dx}=0$, equivalently $\frac{dy}{dt}=0$), CHECK that the $t$-value at that turning point genuinely lies inside the stated domain — a turning point outside $D$ belongs to a part of the curve that was never drawn — and only then combine that interior value with the two endpoint values to state the complete range. This is also the precise reason domain and range are never interchangeable answers to each other: domain is a statement about which INPUTS are used, fixed entirely by where $t$ is allowed to range; range is a statement about which OUTPUTS actually occur, which depends on the full behaviour of $g$ across that domain, turning points included — a structurally different question, and "gave answers in terms of x rather than y" (Jan 2021, Q4) is exactly what happens when a range question is answered with a domain-shaped answer instead. Two further cases complete the picture, both turning on whether $D$ is closed and bounded or not. FIRST, domain-finding when $D$ is open and/or unbounded: there is no literal endpoint value of $t$ to substitute at all — $t>0$ excludes its own lower end and has no upper one — so the closed-bounded shortcut above does not apply directly; each excluded or infinite end has to be reasoned about as a LIMIT instead, asking what $x$ approaches as $t$ tends to the excluded value, and what $x$ approaches (or does) as $t\to\infty$, then combining the two limiting values into an open and/or unbounded $x$-domain (e.g. $x>k$, never $x\geq k$, whenever the limiting value $k$ itself is never actually reached by any real value of $t$). SECOND, range-finding when $g$, rewritten as a function of $x$, is monotonic and has NO interior turning point at all: the completing-the-square method above assumes a turning point exists to find, and searching for one that does not leaves no method whatsoever for the boundary values a monotonic function's range still genuinely needs. The fix here checks two different kinds of boundary directly instead: evaluate the cartesian function AT its own excluded endpoint — valid whenever the function is continuous there, even though that exact $x$-value is itself excluded from the stated domain, because the value is still the genuine limit the range approaches — and separately find the value the function approaches as $x\to\infty$ (for a rational function $\frac{ax+b}{cx+d}$, this second boundary is the ratio of leading coefficients, $\frac{a}{c}$, since the constants $b$ and $d$ become negligible once $x$ is large enough). Both boundaries can equally be reached by working with $t$ directly instead of $x$ — taking the very same two limits of $t$ that gave the open/unbounded $x$-domain above, and substituting them into $y(t)$ rather than into $g(x)$.

Traps — 6

guessed-target-form-instead-of-deriving-it

Confirmed directly: "A small minority assumed the general form of the answer, substituted for x in terms of t and compared this to the expression for y but this strategy was often unsuccessful" (Jan 2021, Q4). The trap is specifically INFORMAL, ungoverned guessing and pattern-matching — trying values, or eyeballing a match, without ever setting up a genuine algebraic identity and solving it for the unknowns. It is NOT the same as assuming a general form outright: this very question's own real mark scheme credits a fully valid, full-marks "Alternative 1" built on exactly that starting move — assume $g(x)=\frac{ax+b}{cx+d}$ (the form the question itself gives), substitute $x=\frac1t+2$ into it, and derive $a,b,c,d$ by equating the result to $y=\frac{1-2t}{3+t}$ as an identity in $t$ (matching coefficients of the powers of $t$ on each side), verbatim: "M1: Assume g(x)=(ax+b)/(cx+d) and substitute in x=1/t+2 … A1: g(x)=(a+(b+2a)t)/(c+(d+2c)t) … A1(M1 on EPEN): Correct numerator or denominator … A1: y=(x-4)/(3x-5)" (WMA14/01, January 2021, Q4(a) mark scheme — see this lesson's own real worked chain above, which reaches the same $g(x)=(x-4)/(3x-5)$ by the other valid route). The distinction that actually matters: deriving the unknowns by a genuine coefficient-matching identity is a legitimate forward derivation; guessing them and checking whether they happen to fit a couple of points is not.

endpoint-substitution-attempted-on-an-open-or-unbounded-domain

The lesson's closed-bounded domain method (substitute the parameter's two literal endpoint values directly) has no endpoint to substitute at all when the parameter's domain is open and/or unbounded — e.g. $t>0$, which excludes $t=0$ and has no upper bound. Confirmed on the real anchor this lesson's worked chain above is built on: the real mark scheme states the domain result plainly — "k = 2 or x > 2" (WMA14/01, January 2021, Q4(a) mark scheme) — reached only by reasoning about limits at each excluded/unbounded end of t, never by substituting a literal value of t that does not exist. Confusing the two domain shapes — reaching for direct substitution on an open or unbounded interval, the way the closed-bounded case correctly allows — leaves no method for the domain at all.

domain-confused-with-range

Confirmed on the same question: "Candidates find the concept of domain and range difficult and it was clear that some confused range with domain and gave answers in terms of x rather than y" (Jan 2021, Q4). Domain is a statement about which x-values are used, fixed entirely by the parameter's own restriction; range is a statement about which y-values actually occur, and depends on the full behaviour of y over that domain. Asked for one, giving the other is not a smaller version of a correct answer — it answers a different question.

partial-rearrangement-leaves-t-in-the-equation

Confirmed directly: "A disappointing number of responses ended up with y in terms of both x and t as they only partially rearranged" (Oct 2022, Q1). Eliminating the parameter is only finished once t no longer appears ANYWHERE in the final equation — substituting into one occurrence of t while leaving another untouched produces an expression that looks like progress but is not yet a cartesian equation at all.

calculus-reached-for-on-an-elimination-question

Confirmed on the same question: "A few candidates thought it appropriate to use calculus and scored no marks" (Oct 2022, Q1). Converting between parametric and cartesian form is pure algebraic substitution; differentiating produces a gradient function that still depends on t, or that is simply not the relationship between x and y the question asked for — a plausible-looking wrong tool for this specific question type, not a shortcut to the right one.

endpoints-substituted-for-the-whole-range

Confirmed directly: "Common incorrect solutions followed attempts to substitute either end of the domain in the parametric equation for y. This usually resulted in only one of the two marks being scored, with the minimum value missing" (Oct 2021, Q5(c)). Two endpoint values only describe the two ends of the parameter's domain — they say nothing about an interior turning point, which is exactly where the true minimum (or maximum) of a non-monotonic y(t) can sit, unannounced, between them.

The card

Distinct linear factors (ax+b)(cx+d)(ex+f): one term each, found by cover-up (substitute each factor's own root).
Repeated factor (cx+d)²: TWO terms — B/(cx+d) + C/(cx+d)² — never one. A denominator of degree n needs n unknowns total.
Numerator degree ≥ denominator degree: divide first, or use the identity method with an added constant — A/(x-a)+B/(x-b) alone can't represent it.
∫1/(ax+b)dx = (1/a)ln|ax+b| + c. The 1/a is real and load-bearing — confirmed the single most-documented slip in this topic, in TWO series.
∫1/(ax+b)ⁿdx for n≥2 is POWER-form, no logarithm at all: -1/[(n-1)a(ax+b)ⁿ⁻¹] + c.
x/(x²+5)-style integrands are NOT partial fractions (quadratic factors excluded by spec 2.1) — they integrate by recognition instead.
Expanding a decomposition in x (spec 4.1): factor a constant out of EVERY term to reach k(1+u)ⁿ, same as for a single bracket. The combined range of validity is the SMALLEST of the individual |u|<1 ranges, not the largest.

Why it works — Why a partial-fraction identity can be checked at chosen values of x — and why matching coefficients works too

A partial-fraction decomposition is written with the symbol ≡, not =, and that distinction is the entire justification for both standard solving methods. An equation like $2x=6$ is only true for one value of $x$; an IDENTITY like $8x^2+8x-3\equiv A(x+1)(2x+1)+B(x-2)(2x+1)+C(x-2)(x+1)$ is claimed to be true for every single value of $x$ at once — which means it is safe to substitute ANY value into it, including a value that makes one of the original fraction's own denominators zero (the identity itself, once cleared of denominators, is an ordinary polynomial equation defined everywhere, even where the original fraction wasn't). Substituting the root of one bracket makes every OTHER bracket's term vanish on the right-hand side, leaving an equation in only one unknown — this is the cover-up rule, and it works precisely because the identity is guaranteed to hold at that value too, not despite it. A second, completely independent consequence of the same ≡ symbol: two polynomials that are equal for every value of $x$ must have identical coefficients of every power of $x$ — $x^2$, $x^1$, $x^0$ separately — since a polynomial is entirely determined by its coefficients. This is why comparing coefficients is an equally valid, equally rigorous route to the same unknowns, not an approximation or a shortcut: both methods are reading off different, but equally true, consequences of one ≡ statement.

Traps — 6

extra-power-created-by-over-multiplying-a-repeated-factor

A real examiner report on a repeated-linear-factor decomposition (a DIFFERENT question from this lesson's own (3x+1)(x−2)² example, with its own denominator built around a bracket of (2x+1)) records: "Relatively few students made the error of multiplying by the product of the three denominators so having (2x+1)³" (Jan 2024, Q2(a)) — i.e. treating the repeated bracket as though it needed multiplying by itself an extra, unneeded time when clearing denominators, turning a correct squared bracket into a wrong cubed one. Real, but a genuine minority error ("relatively few"), not the dominant failure mode on this question type — that distinction is worth keeping honest rather than inflating.

simultaneous-equations-more-error-prone-than-substitution

A real examiner report on the same question records a genuine, quotable method-choice recommendation: simultaneous equations built from expanded coefficients, used by a significant number of candidates, were "more prone to error than attempts via substitution" (Jan 2024, Q2(a)) — exactly the finding this lesson's own method-comparison block anchors on. Both methods are legitimate and both reach full marks; the real record's point is which one real scripts actually get wrong more often, not that one method is invalid.

sign-lost-in-a-long-division-remainder

On a question comparing the identity method against long division as two legitimate routes to the same decomposition: "The most common error with candidates who used this method was to set the numerator of the partial fraction equal to 6 rather than -6 and then forgetting the negative when the values were put back into the expression" (Oct 2021, Q3(a)) — a sign lost while forming the remainder during division, then lost a SECOND time when substituting values back in. This lesson's own marked-solution above recreates the same failure class with its own numbers (2x+7 corrupted to 2x−1), not the real question's own 6/−6.

1-over-a-scaling-factor-omitted-on-a-ln-integral

The single most robustly documented error in this whole topic — confirmed independently in TWO separate series: "a common error was to omit the ½ in ∫1/(2x−1) dx = ½ ln(2x−1)" (Oct 2022, Q2), and again, on a different question testing the same skill: "A number of students failed to divide the second logarithm by two" (Jan 2024, Q2(b)). Both quotes describe the identical omission — the reciprocal of the x-coefficient inside the bracket, dropped from the front of the logarithm — this lesson's own marked-solution above is built specifically to anchor on this two-series-confirmed finding.

ln-rule-misapplied-to-a-repeated-factor

Flagged honestly as VERIDIAN-original pedagogical inference, NOT a sourced examiner-report finding — no quote in this unit's research bank documents this specific confusion. The two integrand shapes $\frac{1}{ax+b}$ and $\frac{1}{(ax+b)^2}$ look almost identical on the page, and it would be a natural mistake to apply the ln rule to BOTH, writing something like $\ln|(x-2)^2|$ for the squared case instead of the correct power-rule answer $\frac{-1}{x-2}$. Included here because the two integrand shapes really are this visually close, and this lesson's own mechanism block above exists specifically to derive why they need genuinely different rules — not because a real report has been found documenting students making exactly this error on this paper.

degree-condition-overlooked-before-attempting-the-basic-form

Anchored to spec 2.1's own explicit clause — "The degree of the numerator may equal or exceed the degree of the denominator" — rather than to a specific examiner-report quote describing students missing this: no citation in this unit's research bank documents this exact failure in these terms. Attempting $\frac{A}{x-a}+\frac{B}{x-b}$ directly on a fraction whose numerator's degree already matches the denominator's leads nowhere, because no choice of A and B can make that form reproduce a fraction that doesn't tend to 0 as x grows large. Worth checking degree BEFORE choosing a decomposition form, not after the first attempt fails.

The card

Distinct linear factors (ax+b)(cx+d)(ex+f): one term each, found by cover-up (substitute each factor's own root).
Repeated factor (cx+d)²: TWO terms — B/(cx+d) + C/(cx+d)² — never one. A denominator of degree n needs n unknowns total.
Numerator degree ≥ denominator degree: divide first, or use the identity method with an added constant — A/(x-a)+B/(x-b) alone can't represent it.
∫1/(ax+b)dx = (1/a)ln|ax+b| + c. The 1/a is real and load-bearing — confirmed the single most-documented slip in this topic, in TWO series.
∫1/(ax+b)ⁿdx for n≥2 is POWER-form, no logarithm at all: -1/[(n-1)a(ax+b)ⁿ⁻¹] + c.
x/(x²+5)-style integrands are NOT partial fractions (quadratic factors excluded by spec 2.1) — they integrate by recognition instead.
Expanding a decomposition in x (spec 4.1): factor a constant out of EVERY term to reach k(1+u)ⁿ, same as for a single bracket. The combined range of validity is the SMALLEST of the individual |u|<1 ranges, not the largest.

Why it works — Why a partial-fraction identity can be checked at chosen values of x — and why matching coefficients works too

A partial-fraction decomposition is written with the symbol ≡, not =, and that distinction is the entire justification for both standard solving methods. An equation like $2x=6$ is only true for one value of $x$; an IDENTITY like $8x^2+8x-3\equiv A(x+1)(2x+1)+B(x-2)(2x+1)+C(x-2)(x+1)$ is claimed to be true for every single value of $x$ at once — which means it is safe to substitute ANY value into it, including a value that makes one of the original fraction's own denominators zero (the identity itself, once cleared of denominators, is an ordinary polynomial equation defined everywhere, even where the original fraction wasn't). Substituting the root of one bracket makes every OTHER bracket's term vanish on the right-hand side, leaving an equation in only one unknown — this is the cover-up rule, and it works precisely because the identity is guaranteed to hold at that value too, not despite it. A second, completely independent consequence of the same ≡ symbol: two polynomials that are equal for every value of $x$ must have identical coefficients of every power of $x$ — $x^2$, $x^1$, $x^0$ separately — since a polynomial is entirely determined by its coefficients. This is why comparing coefficients is an equally valid, equally rigorous route to the same unknowns, not an approximation or a shortcut: both methods are reading off different, but equally true, consequences of one ≡ statement.

Traps — 6

extra-power-created-by-over-multiplying-a-repeated-factor

A real examiner report on a repeated-linear-factor decomposition (a DIFFERENT question from this lesson's own (3x+1)(x−2)² example, with its own denominator built around a bracket of (2x+1)) records: "Relatively few students made the error of multiplying by the product of the three denominators so having (2x+1)³" (Jan 2024, Q2(a)) — i.e. treating the repeated bracket as though it needed multiplying by itself an extra, unneeded time when clearing denominators, turning a correct squared bracket into a wrong cubed one. Real, but a genuine minority error ("relatively few"), not the dominant failure mode on this question type — that distinction is worth keeping honest rather than inflating.

simultaneous-equations-more-error-prone-than-substitution

A real examiner report on the same question records a genuine, quotable method-choice recommendation: simultaneous equations built from expanded coefficients, used by a significant number of candidates, were "more prone to error than attempts via substitution" (Jan 2024, Q2(a)) — exactly the finding this lesson's own method-comparison block anchors on. Both methods are legitimate and both reach full marks; the real record's point is which one real scripts actually get wrong more often, not that one method is invalid.

sign-lost-in-a-long-division-remainder

On a question comparing the identity method against long division as two legitimate routes to the same decomposition: "The most common error with candidates who used this method was to set the numerator of the partial fraction equal to 6 rather than -6 and then forgetting the negative when the values were put back into the expression" (Oct 2021, Q3(a)) — a sign lost while forming the remainder during division, then lost a SECOND time when substituting values back in. This lesson's own marked-solution above recreates the same failure class with its own numbers (2x+7 corrupted to 2x−1), not the real question's own 6/−6.

1-over-a-scaling-factor-omitted-on-a-ln-integral

The single most robustly documented error in this whole topic — confirmed independently in TWO separate series: "a common error was to omit the ½ in ∫1/(2x−1) dx = ½ ln(2x−1)" (Oct 2022, Q2), and again, on a different question testing the same skill: "A number of students failed to divide the second logarithm by two" (Jan 2024, Q2(b)). Both quotes describe the identical omission — the reciprocal of the x-coefficient inside the bracket, dropped from the front of the logarithm — this lesson's own marked-solution above is built specifically to anchor on this two-series-confirmed finding.

ln-rule-misapplied-to-a-repeated-factor

Flagged honestly as VERIDIAN-original pedagogical inference, NOT a sourced examiner-report finding — no quote in this unit's research bank documents this specific confusion. The two integrand shapes $\frac{1}{ax+b}$ and $\frac{1}{(ax+b)^2}$ look almost identical on the page, and it would be a natural mistake to apply the ln rule to BOTH, writing something like $\ln|(x-2)^2|$ for the squared case instead of the correct power-rule answer $\frac{-1}{x-2}$. Included here because the two integrand shapes really are this visually close, and this lesson's own mechanism block above exists specifically to derive why they need genuinely different rules — not because a real report has been found documenting students making exactly this error on this paper.

degree-condition-overlooked-before-attempting-the-basic-form

Anchored to spec 2.1's own explicit clause — "The degree of the numerator may equal or exceed the degree of the denominator" — rather than to a specific examiner-report quote describing students missing this: no citation in this unit's research bank documents this exact failure in these terms. Attempting $\frac{A}{x-a}+\frac{B}{x-b}$ directly on a fraction whose numerator's degree already matches the denominator's leads nowhere, because no choice of A and B can make that form reproduce a fraction that doesn't tend to 0 as x grows large. Worth checking degree BEFORE choosing a decomposition form, not after the first attempt fails.

The card

Implicit: differentiate every term w.r.t. x. Y-terms pick up dy/dx (chain rule); x-only terms don't. Mixed xy-terms need the product rule first.
Parametric: dy/dx = (dy/dt) ÷ (dx/dt), where dx/dt ≠ 0 — a division of the two rates, never a product or a swapped ratio.
Tangent/normal at P: substitute P's OWN coordinates into dy/dx (implicit: both its x and y). Using (0,0) by default has cost real candidates every mark, even after finding P correctly.
Perpendicular gradients: m(normal) = −1/m(tangent). Line: y − y₁ = m(x − x₁).
xy² needs product AND chain rule together: d/dx(xy²) = 2xy·dy/dx + y² — two terms, the hardest single step documented in this topic.
(x+y)ⁿ: chain rule directly, or expand and differentiate term by term — both routes credited equally. Simplify a compound parametric fraction carefully: a real script inverted 1/(3sin²t) into 3cos²t.

Why it works — Where the dy/dx factor comes from — the chain rule, applied to y itself

Recall the chain rule from wma13-differentiation-rules: $\frac{d}{dx}\left[(g(x))^n\right] = n(g(x))^{n-1} g'(x)$ — differentiate the power, then multiply by the derivative of whatever is inside it. Now substitute $g(x) = y$, treating y explicitly as a function of x, however unknown its formula is: $\frac{d}{dx}\left[y^n\right] = ny^{n-1}\frac{dy}{dx}$. That is the entire rule. Nothing new has been introduced — it is the same chain rule, with the "inner function" being y itself, and the same reasoning extends to any function of y alone: $\frac{d}{dx}[\sin y] = \cos y \cdot \frac{dy}{dx}$, $\frac{d}{dx}[e^y] = e^y\frac{dy}{dx}$, and so on, term by term. A term that mixes x and y together — $x^2y$, $xy^2$ — is not a pure function of y alone, so this rule cannot be applied to the whole term directly: it is a PRODUCT of an x-part and a y-part, and needs the product rule first, with the chain rule then firing only on whichever factor actually contains y. This exact compound situation — product rule and chain rule together, in one term — is named directly in the verified record as the single hardest mechanical step in implicit differentiation across the whole archive reviewed (Jan 2024, Q3(a)), and the worked chain immediately below exists to take it apart in full. One more consequence worth stating before it gets used: a term in x ALONE differentiates with no dy/dx anywhere, because $\frac{d}{dx}(x) = 1$ — the chain-rule factor for x differentiating with respect to x is invisible, being exactly 1, which is also exactly why a bare constant (with no y and no x at all) differentiates to 0 with nothing attached to it. Reflexively pattern-matching "y-terms get a dy/dx" onto a term that never had a y in it — a genuinely documented error, examined directly in this lesson's own trap taxonomy below (Oct 2021, Q1) — comes from treating this as a rule to apply on sight, rather than as something the chain rule only ever does to a term that actually contains y.

Traps — 5

wrong-point-used-for-gradient-evaluation

Confirmed directly, and the single most consequential documented trap in this whole topic because of how little it costs to avoid and how much it costs to fall into: on a question asking for the tangent or normal at a named point P, "there was a surprisingly large number of candidates who simply took P to be the origin and lost all marks in this part. In some cases, the candidate found the correct co-ordinates for P but still used the point (0, 0) to evaluate the gradient" (Jan 2021, Q6). The second sentence is the one worth reading twice: the error is not in finding P, and not in differentiating — it is entirely in which point gets substituted into an otherwise-correct dy/dx expression, and it is documented as costing every mark in the part, not a partial share of it.

product-rule-term-dropped-differentiating-a-mixed-xy-term

One of three specific, individually-quotable slip patterns confirmed on the SAME real question (Oct 2021, Q1 — one question documenting three distinct errors at once, not three separate series): "differentiating 3x²y → 6x dy/dx." The correct result is 6xy + 3x²·dy/dx — two terms, from the product rule applied to a product of an x-part and a y-part. The documented wrong answer keeps only a differentiated x²-coefficient (6x) and tacks a bare dy/dx onto it, which is what happens when the product rule is skipped entirely and the term is treated as though only ONE of its two factors depended on x.

chain-rule-y-factor-dropped-differentiating-a-y-power-term

A second pattern from the same real question and series (Oct 2021, Q1): "differentiating 4y² → 8 dy/dx." The correct result is 8y·dy/dx — the chain rule multiplies by the derivative of y² with respect to y (which is 2y, not 1) AND by dy/dx. The documented wrong answer keeps the dy/dx factor but drops the leftover y entirely, as though d/dy(y²) were 1 instead of 2y — the mirror-image gap to the trap above: that one drops the product rule's other factor; this one drops half of the chain rule's own contribution.

dy-dx-added-reflexively-to-a-term-with-no-y-at-all

The third pattern from the same question, and the one WMA14-verified-facts.md itself calls "surprising": "differentiating 4x² + 8 → 8x + 8 dy/dx" (Oct 2021, Q1). The correct result is simply 8x — a constant, 8, differentiates to 0 regardless of what else is in the equation, and there is no y anywhere in this expression for a chain rule to attach to. This is an OVER-application of the "y-terms pick up dy/dx" rule from this lesson's own mechanism above: taught as a pattern to spot rather than derived from what the chain rule actually does, it gets pattern-matched onto a term that never had a y in it at all.

reciprocal-error-simplifying-a-compound-parametric-fraction

Confirmed directly on a real parametric-tangent question: a compound dy/dx fraction was seen simplified as "1/(3sin²t) → 3cos²t" (Oct 2022, Q6) — inverting the fraction and substituting cosine for sine in the same move, after both dy/dt and dx/dt had already been found and combined correctly. This is the documented cost of treating "simplify the fraction" as a mechanical afterthought once the calculus is done: the calculus in this case was already finished and correct, and the mark was lost on the algebra that came after it.

The card

|v| = √(x²+y²+z²) — 3D Pythagoras. Unit vector in the direction of v: v/|v| (check: its own components' squares sum to 1).
AB = OB − OA = b − a (destination minus start — NOT a − b, which gives BA, the reverse).
Addition/subtraction: component-wise. Scalar k·v: scales magnitude by |k|; reverses direction if k < 0.
Distance between (x₁,y₁,z₁) and (x₂,y₂,z₂): d² = (x₁−x₂)²+(y₁−y₂)²+(z₁−z₂)² — literally |AB|² written out in coordinates.
Foot of perpendicular from A to line r = c + t·d: form AX = X − a first (X = c + t·d), then solve AX·d = 0 for t. NOT the raw general point X dotted with d instead of AX — (c+t·d)·d = 0 still solves, just for the wrong point; skipping the subtraction of a is the error, not unsolvability.
Shortest distance from A to the line = |AX| at that t. Same t also minimises |AX|² — check via differentiation or completing the square.
Reflection of A in the line: X is the midpoint of A and its image B, so B = 2X − A (spec 7.4's identity, reused rather than a new formula).

Why it works — Where |v| = √(x² + y² + z²) actually comes from — Pythagoras, run twice

This formula is not a new rule to memorise; it is ordinary 2D Pythagoras, applied twice. Take $\mathbf{v}=(x,y,z)$. First, project $\mathbf{v}$ straight down onto the $xy$-plane — this projection is the vector $(x,y,0)$, and by the familiar 2D case, its length is $\sqrt{x^2+y^2}$. Second, notice that $\mathbf{v}$ itself, its projection $(x,y,0)$, and a vertical segment of length $|z|$ (running straight up or down from the projection's endpoint to $\mathbf{v}$'s own endpoint) form a second right-angled triangle — right-angled specifically because the $z$-axis is perpendicular to the entire $xy$-plane, so it is perpendicular to the projection no matter which direction the projection itself points in. Applying Pythagoras to THIS triangle: $|\mathbf{v}|^2 = \left(\sqrt{x^2+y^2}\right)^2 + z^2 = x^2+y^2+z^2$, so $|\mathbf{v}| = \sqrt{x^2+y^2+z^2}$ — the 3D magnitude formula, reached by two applications of a 2D fact already known, rather than handed down as something new about three dimensions specifically.

Traps — 4

general-point-substituted-for-ax

Confirmed directly on exactly this question type, and more precisely than "dots the direction vector with itself" would suggest: the real, examiner-report-documented error is not dotting the direction vector with itself (that equation has no t in it and is trivially, visibly unsolvable) — it is skipping the step that forms AX = (general point) − a, and instead dotting the raw general point on the line straight against d: "most incorrect responses attempted to set [the general point on the line] . [the direction vector] = 0 rather than AX . [the direction vector] = 0... This immediately resulted in 0 marks for (a)(i)" (Oct 2021, Q7(a)). This equation genuinely DOES contain t and is solvable — on the real exam question it gives a specific, plausible-looking wrong value ($\lambda=\tfrac{37}{50}$, versus the correct $\lambda=\tfrac{2}{5}$) — which is exactly why it is dangerous: nothing about the arithmetic looks broken, only the choice of vector is wrong. Contrast this with genuinely dotting the direction vector with itself, d·d = 0: that IS structurally unsolvable (no t at all) — a different, much rarer, and merely hypothetical slip, not the one this examiner report documents. This is the single documented error this section of the facts bank contains for this question type; a real, single-source citation, not padded into a broader pattern.

ab-formed-in-the-wrong-order

A general, VERIDIAN-identified trap rather than a sourced one: forming AB as a − b instead of b − a gives BA (every component's sign reversed), the reverse of what was actually asked for. Unlike the distance formula, which is symmetric because it squares each difference (so it doesn't matter which point is subtracted from which), the vector AB is NOT symmetric — getting the order backwards produces a genuinely wrong vector, not merely a differently-labelled correct one.

magnitude-missing-the-final-square-root

A general, VERIDIAN-identified trap: stopping at the sum of squares (e.g. reporting 169 instead of 13) rather than taking the final square root. This is a genuinely easy line to skip, since every individual term along the way — each component squared, then summed — was computed correctly; the omission is the very last step, not an error anywhere inside the working.

unit-vector-not-actually-normalised

A general, VERIDIAN-identified trap: giving the original vector itself, or a partially-scaled version of it, as "the unit vector," rather than dividing every one of its components by its own magnitude. The fast check named earlier in this lesson — do the squares of the claimed answer's own components sum to exactly 1? — catches this immediately, and is worth running as a habit rather than only when something already looks suspicious.

The card

|v| = √(x²+y²+z²) — 3D Pythagoras. Unit vector in the direction of v: v/|v| (check: its own components' squares sum to 1).
AB = OB − OA = b − a (destination minus start — NOT a − b, which gives BA, the reverse).
Addition/subtraction: component-wise. Scalar k·v: scales magnitude by |k|; reverses direction if k < 0.
Distance between (x₁,y₁,z₁) and (x₂,y₂,z₂): d² = (x₁−x₂)²+(y₁−y₂)²+(z₁−z₂)² — literally |AB|² written out in coordinates.
Foot of perpendicular from A to line r = c + t·d: form AX = X − a first (X = c + t·d), then solve AX·d = 0 for t. NOT the raw general point X dotted with d instead of AX — (c+t·d)·d = 0 still solves, just for the wrong point; skipping the subtraction of a is the error, not unsolvability.
Shortest distance from A to the line = |AX| at that t. Same t also minimises |AX|² — check via differentiation or completing the square.
Reflection of A in the line: X is the midpoint of A and its image B, so B = 2X − A (spec 7.4's identity, reused rather than a new formula).

Why it works — Where |v| = √(x² + y² + z²) actually comes from — Pythagoras, run twice

This formula is not a new rule to memorise; it is ordinary 2D Pythagoras, applied twice. Take $\mathbf{v}=(x,y,z)$. First, project $\mathbf{v}$ straight down onto the $xy$-plane — this projection is the vector $(x,y,0)$, and by the familiar 2D case, its length is $\sqrt{x^2+y^2}$. Second, notice that $\mathbf{v}$ itself, its projection $(x,y,0)$, and a vertical segment of length $|z|$ (running straight up or down from the projection's endpoint to $\mathbf{v}$'s own endpoint) form a second right-angled triangle — right-angled specifically because the $z$-axis is perpendicular to the entire $xy$-plane, so it is perpendicular to the projection no matter which direction the projection itself points in. Applying Pythagoras to THIS triangle: $|\mathbf{v}|^2 = \left(\sqrt{x^2+y^2}\right)^2 + z^2 = x^2+y^2+z^2$, so $|\mathbf{v}| = \sqrt{x^2+y^2+z^2}$ — the 3D magnitude formula, reached by two applications of a 2D fact already known, rather than handed down as something new about three dimensions specifically.

Traps — 4

general-point-substituted-for-ax

Confirmed directly on exactly this question type, and more precisely than "dots the direction vector with itself" would suggest: the real, examiner-report-documented error is not dotting the direction vector with itself (that equation has no t in it and is trivially, visibly unsolvable) — it is skipping the step that forms AX = (general point) − a, and instead dotting the raw general point on the line straight against d: "most incorrect responses attempted to set [the general point on the line] . [the direction vector] = 0 rather than AX . [the direction vector] = 0... This immediately resulted in 0 marks for (a)(i)" (Oct 2021, Q7(a)). This equation genuinely DOES contain t and is solvable — on the real exam question it gives a specific, plausible-looking wrong value ($\lambda=\tfrac{37}{50}$, versus the correct $\lambda=\tfrac{2}{5}$) — which is exactly why it is dangerous: nothing about the arithmetic looks broken, only the choice of vector is wrong. Contrast this with genuinely dotting the direction vector with itself, d·d = 0: that IS structurally unsolvable (no t at all) — a different, much rarer, and merely hypothetical slip, not the one this examiner report documents. This is the single documented error this section of the facts bank contains for this question type; a real, single-source citation, not padded into a broader pattern.

ab-formed-in-the-wrong-order

A general, VERIDIAN-identified trap rather than a sourced one: forming AB as a − b instead of b − a gives BA (every component's sign reversed), the reverse of what was actually asked for. Unlike the distance formula, which is symmetric because it squares each difference (so it doesn't matter which point is subtracted from which), the vector AB is NOT symmetric — getting the order backwards produces a genuinely wrong vector, not merely a differently-labelled correct one.

magnitude-missing-the-final-square-root

A general, VERIDIAN-identified trap: stopping at the sum of squares (e.g. reporting 169 instead of 13) rather than taking the final square root. This is a genuinely easy line to skip, since every individual term along the way — each component squared, then summed — was computed correctly; the omission is the very last step, not an error anywhere inside the working.

unit-vector-not-actually-normalised

A general, VERIDIAN-identified trap: giving the original vector itself, or a partially-scaled version of it, as "the unit vector," rather than dividing every one of its components by its own magnitude. The fast check named earlier in this lesson — do the squares of the claimed answer's own components sum to exactly 1? — catches this immediately, and is worth running as a habit rather than only when something already looks suspicious.

The card

|v| = √(x²+y²+z²) — 3D Pythagoras. Unit vector in the direction of v: v/|v| (check: its own components' squares sum to 1).
AB = OB − OA = b − a (destination minus start — NOT a − b, which gives BA, the reverse).
Addition/subtraction: component-wise. Scalar k·v: scales magnitude by |k|; reverses direction if k < 0.
Distance between (x₁,y₁,z₁) and (x₂,y₂,z₂): d² = (x₁−x₂)²+(y₁−y₂)²+(z₁−z₂)² — literally |AB|² written out in coordinates.
Foot of perpendicular from A to line r = c + t·d: form AX = X − a first (X = c + t·d), then solve AX·d = 0 for t. NOT the raw general point X dotted with d instead of AX — (c+t·d)·d = 0 still solves, just for the wrong point; skipping the subtraction of a is the error, not unsolvability.
Shortest distance from A to the line = |AX| at that t. Same t also minimises |AX|² — check via differentiation or completing the square.
Reflection of A in the line: X is the midpoint of A and its image B, so B = 2X − A (spec 7.4's identity, reused rather than a new formula).

Why it works — Where |v| = √(x² + y² + z²) actually comes from — Pythagoras, run twice

This formula is not a new rule to memorise; it is ordinary 2D Pythagoras, applied twice. Take $\mathbf{v}=(x,y,z)$. First, project $\mathbf{v}$ straight down onto the $xy$-plane — this projection is the vector $(x,y,0)$, and by the familiar 2D case, its length is $\sqrt{x^2+y^2}$. Second, notice that $\mathbf{v}$ itself, its projection $(x,y,0)$, and a vertical segment of length $|z|$ (running straight up or down from the projection's endpoint to $\mathbf{v}$'s own endpoint) form a second right-angled triangle — right-angled specifically because the $z$-axis is perpendicular to the entire $xy$-plane, so it is perpendicular to the projection no matter which direction the projection itself points in. Applying Pythagoras to THIS triangle: $|\mathbf{v}|^2 = \left(\sqrt{x^2+y^2}\right)^2 + z^2 = x^2+y^2+z^2$, so $|\mathbf{v}| = \sqrt{x^2+y^2+z^2}$ — the 3D magnitude formula, reached by two applications of a 2D fact already known, rather than handed down as something new about three dimensions specifically.

Traps — 4

general-point-substituted-for-ax

Confirmed directly on exactly this question type, and more precisely than "dots the direction vector with itself" would suggest: the real, examiner-report-documented error is not dotting the direction vector with itself (that equation has no t in it and is trivially, visibly unsolvable) — it is skipping the step that forms AX = (general point) − a, and instead dotting the raw general point on the line straight against d: "most incorrect responses attempted to set [the general point on the line] . [the direction vector] = 0 rather than AX . [the direction vector] = 0... This immediately resulted in 0 marks for (a)(i)" (Oct 2021, Q7(a)). This equation genuinely DOES contain t and is solvable — on the real exam question it gives a specific, plausible-looking wrong value ($\lambda=\tfrac{37}{50}$, versus the correct $\lambda=\tfrac{2}{5}$) — which is exactly why it is dangerous: nothing about the arithmetic looks broken, only the choice of vector is wrong. Contrast this with genuinely dotting the direction vector with itself, d·d = 0: that IS structurally unsolvable (no t at all) — a different, much rarer, and merely hypothetical slip, not the one this examiner report documents. This is the single documented error this section of the facts bank contains for this question type; a real, single-source citation, not padded into a broader pattern.

ab-formed-in-the-wrong-order

A general, VERIDIAN-identified trap rather than a sourced one: forming AB as a − b instead of b − a gives BA (every component's sign reversed), the reverse of what was actually asked for. Unlike the distance formula, which is symmetric because it squares each difference (so it doesn't matter which point is subtracted from which), the vector AB is NOT symmetric — getting the order backwards produces a genuinely wrong vector, not merely a differently-labelled correct one.

magnitude-missing-the-final-square-root

A general, VERIDIAN-identified trap: stopping at the sum of squares (e.g. reporting 169 instead of 13) rather than taking the final square root. This is a genuinely easy line to skip, since every individual term along the way — each component squared, then summed — was computed correctly; the omission is the very last step, not an error anywhere inside the working.

unit-vector-not-actually-normalised

A general, VERIDIAN-identified trap: giving the original vector itself, or a partially-scaled version of it, as "the unit vector," rather than dividing every one of its components by its own magnitude. The fast check named earlier in this lesson — do the squares of the claimed answer's own components sum to exactly 1? — catches this immediately, and is worth running as a habit rather than only when something already looks suspicious.

The card

|v| = √(x²+y²+z²) — 3D Pythagoras. Unit vector in the direction of v: v/|v| (check: its own components' squares sum to 1).
AB = OB − OA = b − a (destination minus start — NOT a − b, which gives BA, the reverse).
Addition/subtraction: component-wise. Scalar k·v: scales magnitude by |k|; reverses direction if k < 0.
Distance between (x₁,y₁,z₁) and (x₂,y₂,z₂): d² = (x₁−x₂)²+(y₁−y₂)²+(z₁−z₂)² — literally |AB|² written out in coordinates.
Foot of perpendicular from A to line r = c + t·d: form AX = X − a first (X = c + t·d), then solve AX·d = 0 for t. NOT the raw general point X dotted with d instead of AX — (c+t·d)·d = 0 still solves, just for the wrong point; skipping the subtraction of a is the error, not unsolvability.
Shortest distance from A to the line = |AX| at that t. Same t also minimises |AX|² — check via differentiation or completing the square.
Reflection of A in the line: X is the midpoint of A and its image B, so B = 2X − A (spec 7.4's identity, reused rather than a new formula).

Why it works — Where |v| = √(x² + y² + z²) actually comes from — Pythagoras, run twice

This formula is not a new rule to memorise; it is ordinary 2D Pythagoras, applied twice. Take $\mathbf{v}=(x,y,z)$. First, project $\mathbf{v}$ straight down onto the $xy$-plane — this projection is the vector $(x,y,0)$, and by the familiar 2D case, its length is $\sqrt{x^2+y^2}$. Second, notice that $\mathbf{v}$ itself, its projection $(x,y,0)$, and a vertical segment of length $|z|$ (running straight up or down from the projection's endpoint to $\mathbf{v}$'s own endpoint) form a second right-angled triangle — right-angled specifically because the $z$-axis is perpendicular to the entire $xy$-plane, so it is perpendicular to the projection no matter which direction the projection itself points in. Applying Pythagoras to THIS triangle: $|\mathbf{v}|^2 = \left(\sqrt{x^2+y^2}\right)^2 + z^2 = x^2+y^2+z^2$, so $|\mathbf{v}| = \sqrt{x^2+y^2+z^2}$ — the 3D magnitude formula, reached by two applications of a 2D fact already known, rather than handed down as something new about three dimensions specifically.

Traps — 4

general-point-substituted-for-ax

Confirmed directly on exactly this question type, and more precisely than "dots the direction vector with itself" would suggest: the real, examiner-report-documented error is not dotting the direction vector with itself (that equation has no t in it and is trivially, visibly unsolvable) — it is skipping the step that forms AX = (general point) − a, and instead dotting the raw general point on the line straight against d: "most incorrect responses attempted to set [the general point on the line] . [the direction vector] = 0 rather than AX . [the direction vector] = 0... This immediately resulted in 0 marks for (a)(i)" (Oct 2021, Q7(a)). This equation genuinely DOES contain t and is solvable — on the real exam question it gives a specific, plausible-looking wrong value ($\lambda=\tfrac{37}{50}$, versus the correct $\lambda=\tfrac{2}{5}$) — which is exactly why it is dangerous: nothing about the arithmetic looks broken, only the choice of vector is wrong. Contrast this with genuinely dotting the direction vector with itself, d·d = 0: that IS structurally unsolvable (no t at all) — a different, much rarer, and merely hypothetical slip, not the one this examiner report documents. This is the single documented error this section of the facts bank contains for this question type; a real, single-source citation, not padded into a broader pattern.

ab-formed-in-the-wrong-order

A general, VERIDIAN-identified trap rather than a sourced one: forming AB as a − b instead of b − a gives BA (every component's sign reversed), the reverse of what was actually asked for. Unlike the distance formula, which is symmetric because it squares each difference (so it doesn't matter which point is subtracted from which), the vector AB is NOT symmetric — getting the order backwards produces a genuinely wrong vector, not merely a differently-labelled correct one.

magnitude-missing-the-final-square-root

A general, VERIDIAN-identified trap: stopping at the sum of squares (e.g. reporting 169 instead of 13) rather than taking the final square root. This is a genuinely easy line to skip, since every individual term along the way — each component squared, then summed — was computed correctly; the omission is the very last step, not an error anywhere inside the working.

unit-vector-not-actually-normalised

A general, VERIDIAN-identified trap: giving the original vector itself, or a partially-scaled version of it, as "the unit vector," rather than dividing every one of its components by its own magnitude. The fast check named earlier in this lesson — do the squares of the claimed answer's own components sum to exactly 1? — catches this immediately, and is worth running as a habit rather than only when something already looks suspicious.

The card

|v| = √(x²+y²+z²) — 3D Pythagoras. Unit vector in the direction of v: v/|v| (check: its own components' squares sum to 1).
AB = OB − OA = b − a (destination minus start — NOT a − b, which gives BA, the reverse).
Addition/subtraction: component-wise. Scalar k·v: scales magnitude by |k|; reverses direction if k < 0.
Distance between (x₁,y₁,z₁) and (x₂,y₂,z₂): d² = (x₁−x₂)²+(y₁−y₂)²+(z₁−z₂)² — literally |AB|² written out in coordinates.
Foot of perpendicular from A to line r = c + t·d: form AX = X − a first (X = c + t·d), then solve AX·d = 0 for t. NOT the raw general point X dotted with d instead of AX — (c+t·d)·d = 0 still solves, just for the wrong point; skipping the subtraction of a is the error, not unsolvability.
Shortest distance from A to the line = |AX| at that t. Same t also minimises |AX|² — check via differentiation or completing the square.
Reflection of A in the line: X is the midpoint of A and its image B, so B = 2X − A (spec 7.4's identity, reused rather than a new formula).

Why it works — Where |v| = √(x² + y² + z²) actually comes from — Pythagoras, run twice

This formula is not a new rule to memorise; it is ordinary 2D Pythagoras, applied twice. Take $\mathbf{v}=(x,y,z)$. First, project $\mathbf{v}$ straight down onto the $xy$-plane — this projection is the vector $(x,y,0)$, and by the familiar 2D case, its length is $\sqrt{x^2+y^2}$. Second, notice that $\mathbf{v}$ itself, its projection $(x,y,0)$, and a vertical segment of length $|z|$ (running straight up or down from the projection's endpoint to $\mathbf{v}$'s own endpoint) form a second right-angled triangle — right-angled specifically because the $z$-axis is perpendicular to the entire $xy$-plane, so it is perpendicular to the projection no matter which direction the projection itself points in. Applying Pythagoras to THIS triangle: $|\mathbf{v}|^2 = \left(\sqrt{x^2+y^2}\right)^2 + z^2 = x^2+y^2+z^2$, so $|\mathbf{v}| = \sqrt{x^2+y^2+z^2}$ — the 3D magnitude formula, reached by two applications of a 2D fact already known, rather than handed down as something new about three dimensions specifically.

Traps — 4

general-point-substituted-for-ax

Confirmed directly on exactly this question type, and more precisely than "dots the direction vector with itself" would suggest: the real, examiner-report-documented error is not dotting the direction vector with itself (that equation has no t in it and is trivially, visibly unsolvable) — it is skipping the step that forms AX = (general point) − a, and instead dotting the raw general point on the line straight against d: "most incorrect responses attempted to set [the general point on the line] . [the direction vector] = 0 rather than AX . [the direction vector] = 0... This immediately resulted in 0 marks for (a)(i)" (Oct 2021, Q7(a)). This equation genuinely DOES contain t and is solvable — on the real exam question it gives a specific, plausible-looking wrong value ($\lambda=\tfrac{37}{50}$, versus the correct $\lambda=\tfrac{2}{5}$) — which is exactly why it is dangerous: nothing about the arithmetic looks broken, only the choice of vector is wrong. Contrast this with genuinely dotting the direction vector with itself, d·d = 0: that IS structurally unsolvable (no t at all) — a different, much rarer, and merely hypothetical slip, not the one this examiner report documents. This is the single documented error this section of the facts bank contains for this question type; a real, single-source citation, not padded into a broader pattern.

ab-formed-in-the-wrong-order

A general, VERIDIAN-identified trap rather than a sourced one: forming AB as a − b instead of b − a gives BA (every component's sign reversed), the reverse of what was actually asked for. Unlike the distance formula, which is symmetric because it squares each difference (so it doesn't matter which point is subtracted from which), the vector AB is NOT symmetric — getting the order backwards produces a genuinely wrong vector, not merely a differently-labelled correct one.

magnitude-missing-the-final-square-root

A general, VERIDIAN-identified trap: stopping at the sum of squares (e.g. reporting 169 instead of 13) rather than taking the final square root. This is a genuinely easy line to skip, since every individual term along the way — each component squared, then summed — was computed correctly; the omission is the very last step, not an error anywhere inside the working.

unit-vector-not-actually-normalised

A general, VERIDIAN-identified trap: giving the original vector itself, or a partially-scaled version of it, as "the unit vector," rather than dividing every one of its components by its own magnitude. The fast check named earlier in this lesson — do the squares of the claimed answer's own components sum to exactly 1? — catches this immediately, and is worth running as a habit rather than only when something already looks suspicious.

The card

A = ∫y dx = ∫y (dx/dt) dt — the only new step beyond ordinary integration is converting BOTH the dx and the limits into t.
Find dx/dt first. Then solve x(t) = [each given x-value] for t, checking any stated domain (e.g. t ≥ 0) before choosing a root.
Match limits correctly: the t-value for the LOWER x-limit is the lower t-limit; the t-value for the UPPER x-limit is the upper t-limit — even if this reverses the order t itself runs in.
Get the matching right and the sign comes out right automatically — there is no separate rule for 'flip the sign if dx/dt is negative.'
π∫y²(dx/dt)dt is the VOLUME formula (spec 6.1) — a different spec point. Area (spec 6.5) has no π and never squares y.

Why it works — Where A = ∫y (dx/dt) dt actually comes from — the same substitution argument every change of variable uses

The area formula $A=\int_a^b y\,dx$ itself comes from summing thin vertical strips — width $\delta x$, height $y$ — each contributing $y\,\delta x$ to the running total, with the sum becoming an exact integral as $\delta x\to0$. Nothing in that argument cares how $x$ is written down: $x$ could stand on its own, or be written as a function of some other variable $t$, provided a unique $t$ corresponds to each $x$ in the region being swept out (the reason a parameter's range has to be chosen so this holds in the first place). Once $x=x(t)$, the ordinary rule for changing the variable of a definite integral applies exactly as it would anywhere else: $dx=\frac{dx}{dt}\,dt$, and — this is the part that's easy to treat as optional and isn't — the LIMITS have to be re-expressed in terms of $t$ too, because a definite integral's limits are always values of whichever variable its own $d(\ldots)$ names. Leaving the limits as the original x-values while integrating with respect to $t$ doesn't compute the wrong number by some fixable margin; it computes something that isn't the intended integral at all, the same way $\int_a^b y\,\frac{dx}{dt}\,dt$ mixes a $t$-integral with $x$-limits and produces a value with no clean geometric meaning.

Traps — 4

x-limits-used-directly-as-t-limits-without-converting

Reasoned as a direct structural inference from this course's own verified research on the adjacent topic of substitution (spec 6.2, a genuinely different spec point): 'Common reasons for a loss of marks... use of the x limits... instead of the u limits' (Oct 2022, Q7(i), WMA14-verified-facts.md §5.6). Spec 6.5's own technique — $A=\int y\,dx=\int y\,\frac{dx}{dt}\,dt$ — is the identical bookkeeping move: a change of the integral's variable, which always carries the same requirement that the LIMITS change with it. No primary document reviewed for this course names this specific failure on a parametric-area question, because none was found evidencing spec 6.5 at all (see this lesson's own closing flag) — this item is included because the mechanic it's about is structurally identical, not because a report was found saying so.

dx-dt-factor-dropped-entirely

The sister failure to the one above, from the same adjacent substitution research: 'candidates who ignored the dx and simply wrote it as du' (Oct 2021, Q6, WMA14-verified-facts.md §5.6) — the differential-conversion half of exactly the same substitution move spec 6.5 asks for, applied there to a different variable name. On a parametric area question this shows up as integrating $y$ with respect to $t$ directly — $\int y\,dt$ — as though $t$ and $x$ were interchangeable, silently changing the answer by whatever $\frac{dx}{dt}$ actually was. Reasoned from the structure of the technique and the adjacent verified finding, not from a report naming this exact question type.

t-limits-kept-in-the-order-they-were-found-rather-than-the-order-the-x-limits-sit-in

This lesson's own central teaching point, reasoned entirely from the structure of a definite integral rather than from any examiner-report quote — no such quote exists for spec 6.5 in this course's research pass; see the closing flag. $\int_p^q f(t)\,dt=-\int_q^p f(t)\,dt$: swapping the two limits of any definite integral negates it. Whenever $\frac{dx}{dt}$ is negative across the interval used — this lesson's own worked example, $x=4\cos t$, is built specifically to exercise this — the t-value belonging to the SMALLER x-limit is the LARGER t-value, and using the two t-values in whichever order they were found (rather than the order the x-limits they represent actually sit in) produces an area with the correct magnitude and the wrong sign. Checking the sign of $\frac{dx}{dt}$ before writing down the final integral's limits — not after getting a suspicious negative number — is the reliable fix.

area-formula-confused-with-the-adjacent-volume-formula

Reasoned from the two spec points' proximity within the same Integration section (6.1 and 6.5, four items apart in the same numbered list) rather than from any documented confusion between them — no examiner report reviewed comments on this specific mix-up. $A=\int y\,\frac{dx}{dt}\,dt$ (spec 6.5, area) and $V=\pi\int y^2\,\frac{dx}{dt}\,dt$ (spec 6.1's own parametric extension, this lesson's prerequisite) share almost every symbol — the same substitution, the same $\frac{dx}{dt}$ factor, the same limit-conversion step — differing only in whether $y$ is squared and whether $\pi$ is present. Revising both topics in close succession, as their adjacent spec numbering invites, is exactly the situation where the two formulas are most likely to blend into each other.