Technique Masterclasses

Pure Mathematics 1

Deep Dive: The Derivative As a Limit

CODEX Deep-Dive: The Derivative As A Limit (WMA11, spec 4.1)

Provenance (2026-09-29, re-verified same day). Built from a direct read of research/veridian/WMA11-verified-facts.md (§1 spec table of contents; §4.4 Differentiation; §6 "topics with enough verified material"; §7 flagged issues; the full "Lesson-audit log") — grepped specifically for "derivative as a limit", "first principles", "chord", "tends to", "h→0"/"→ 0", "limiting", "Jan 2019 Q6", "Jan 2025 Q5"; the full text of the existing lesson content/veridian/curriculum/wma11/the-derivative-as-a-limit.ts; content/veridian/technique-hub/wma11/codex-formula-and-method-bank.md §1.4.1–1.4.3 (checked for overlap before writing a line of this document); the verbatim spec-guidance block for WMA11 §4 item 1 in content/veridian/specCoverage.ts; and, on this pass's re-verification, signal-topic-brief-the-derivative-as-a-limit.md, which independently caught a citation error in this file's §3 (see that section for the fix). content/veridian/question-bank/wma11.ts was checked and contains no item tagged to this spec point. No new primary-source PDF was opened for this pass beyond re-grepping the facts bank for the corrected citation — this document works from what the facts bank and lesson already established, and adds only what neither of them already states.

0. What this document adds, and what it deliberately does not repeat

The existing lesson (the-derivative-as-a-limit.ts) already carries a full, correct, step-by-step first-principles derivation for f(x) = x² — chord gradient, algebraic cancellation of h, the limit — and states its own sourcing honestly: this is a definitional spec item, not an exam-pattern item, so the lesson's worked numbers are "VERIDIAN-original... not transcribed from a specific real WMA11 exam question." The technique-hub formula bank (§1.4.1) gives the one-line formal statement — f'(x) = lim_{h→0} [f(x+h) − f(x)] / h — and explicitly defers the full derivation to that lesson rather than restating it.

This document does not re-derive x² again. Instead it is additive in three specific ways the lesson and formula bank do not cover:

  1. Other sub-cases of the same limit mechanism — the lesson proves the rule for one function (x²) using one algebraic technique (binomial expansion of a squared bracket). The general power-rule formula in the spec's own memorisation box, f(x) = xⁿ ⟹ f′(x) = nxⁿ⁻¹, is stated for all rational n, and reaching it from the same limit definition for n = 0, 1, 3, −1 and ½ requires three genuinely different algebraic moves, not one. Those are worked below.
  2. The second derivative as the same limit process run twice, worked on a concrete function rather than asserted in the abstract (the lesson states the idea; this document executes it).
  3. A precise, source-grounded synthesis of what WMA11 differentiation questions actually test, as distinct from what spec 4.1 defines — using the facts bank's own §4.4 citations to establish that every verified WMA11 differentiation question found in that pass tests use of an already-obtained f'(x), never the limit-derivation step itself.

Everything below is either a direct algebraic derivation (checkable independently by anyone who runs the algebra) or a claim traceable to WMA11-verified-facts.md, the spec's own text as transcribed there, or specCoverage.ts. Where the real material runs out, each section says so rather than continuing.

1. The spec's exact scope for this item

Spec 4.1, as transcribed and verified against the source PDF (international-a-level-maths-spec.pdf, Specification Issue 3, April 2019, P1.3, pp. 12–16) in WMA11-verified-facts.md §1, covers four things under one item number:

The same guidance line contains one explicit exclusion: "knowledge of the chain rule is not required" at this stage. That boundary is load-bearing for §2 below — it rules out proving the power rule for xⁿ by any method that implicitly relies on the chain rule, and it means a composite function like (2x+1)³ is out of scope for a first-principles treatment at P1.

Note what spec 4.1 is not: it is not spec 4.2 ("Differentiation of xⁿ, and related sums, differences and constant multiples") or 4.3 (tangents and normals as an applied skill). Those are separate lessons in this curriculum and are not re-covered here.

2. The limit mechanism, and why it needs a different algebraic move for each index type

The shared definition, for any function f defined near x = a:

f'(a) = lim_{h→0} [f(a+h) − f(a)] / h

The lesson's own mechanism block makes the one general point that matters for every case below: the division by h must happen — and be algebraically valid — while h is still nonzero, and only after that division is complete does letting h → 0 make sense. The three sub-sections below show that "doing the division validly" is not one trick; it is a different piece of algebra depending on what kind of expression f(a+h) − f(a) turns out to be.

2.1 Positive integer powers: expand, then cancel a common factor of h

This is the lesson's own mechanism for x² (n = 2). The same expand-and-cancel move works for any positive integer n, because (a+h)ⁿ − aⁿ always has a common factor of h once expanded — every term of the expansion except aⁿ itself carries at least one power of h, and the aⁿ term cancels exactly against the −aⁿ.

n = 0 (constant function). f(x) = c for a fixed number c. Then f(a+h) − f(a) = c − c = 0 for every h, so [f(a+h) − f(a)]/h = 0/h = 0 for every h ≠ 0 — no cancellation is even needed, the numerator is already zero before dividing. The limit as h → 0 is 0. f′(x) = 0 for any constant function — matching n·xⁿ⁻¹ at n = 0, since the "function" here is x⁰ = 1 scaled by c.

n = 1 (linear). f(x) = x. [f(a+h) − f(a)]/h = [(a+h) − a]/h = h/h = 1, for every h ≠ 0. The limit is 1. f′(x) = 1, matching n·xⁿ⁻¹ = 1·x⁰ = 1.

n = 3 (cubic). f(x) = x³. (a+h)³ − a³ = 3a²h + 3ah² + h³ (binomial expansion, then the a³ terms cancel). Divide by h: 3a² + 3ah + h², valid for h ≠ 0. Let h → 0: the two h-carrying terms vanish, leaving f′(x) = 3x² — matching n·xⁿ⁻¹ = 3x².

The pattern generalises: for any positive integer n, the binomial expansion of (a+h)ⁿ is aⁿ + n·aⁿ⁻¹h + C(n,2)·aⁿ⁻²h² + … + hⁿ. Subtracting aⁿ and dividing by h leaves n·aⁿ⁻¹ + C(n,2)·aⁿ⁻²h + … + hⁿ⁻¹ — every term after the first still carries a positive power of h, so every one of them vanishes as h → 0, leaving exactly n·aⁿ⁻¹. This is the real reason the spec's memorised formula (f(x) = xⁿ ⟹ f′(x) = nxⁿ⁻¹) holds, but reproducing the general binomial argument is not itself required at P1 — the spec gives students the finished formula to memorise and use (per its own P1.2 formula box) rather than asking them to re-derive it for general n. It is included here as real mathematical background for why the memorised rule is true, not as claimed exam content.

2.2 A negative integer power: combine into a single fraction first, then cancel

Binomial expansion does not apply to a reciprocal — there is no bracket to expand. The algebra that makes the division-by-h valid here is different: combine the two fractions in the numerator into one before anything else happens.

n = −1 (reciprocal). f(x) = 1/x, x ≠ 0.

[f(a+h) − f(a)] / h = [1/(a+h) − 1/a] / h

Combine the numerator over a common denominator a(a+h):

= [(a − (a+h)) / (a(a+h))] / h = [−h / (a(a+h))] / h

Now — and only now — is there a factor of h in both numerator and denominator to cancel, valid for h ≠ 0:

= −1 / (a(a+h))

Let h → 0: f′(x) = −1/x² — matching n·xⁿ⁻¹ at n = −1: (−1)·x⁻² = −1/x². The mechanism here (combine fractions over a common denominator before the h in the denominator can be cancelled) is a distinct algebraic skill from expansion-and-cancel, and it is the one that would actually be tested if a first-principles question on a reciprocal function ever appeared — the spec's own worked-example wording for a related skill, "manipulation of surds (rationalising denominators)" (spec 1.2) and combining algebraic fractions, is the general algebra this sub-case draws on.

2.3 A fractional power: rationalise a surd first, then cancel

Neither expansion nor fraction-combination directly clears the h from a denominator when the function itself is a root. The move here is the one named directly in spec 1.2 — rationalising — applied to a difference of surds.

n = ½ (square root). f(x) = √x, x > 0.

[f(a+h) − f(a)] / h = [√(a+h) − √a] / h

Multiply numerator and denominator by the conjugate expression √(a+h) + √a (valid for any h, since this is multiplying by 1):

= [(a+h) − a] / [h(√(a+h) + √a)] = h / [h(√(a+h) + √a)]

Cancel h, valid for h ≠ 0:

= 1 / [√(a+h) + √a]

Let h → 0: f′(x) = 1/(2√x) — matching n·xⁿ⁻¹ at n = ½: (½)·x⁻¹ᐟ² = 1/(2√x). This is the third distinct mechanism (rationalise, then cancel) needed to reach the general power rule from the same limit definition, and it is the one that connects this spec item directly back to spec 1.2's own surd-manipulation content — a real, checkable cross-topic link, not an invented one.

2.4 What this sub-case survey actually establishes

Three algebraically distinct techniques — expand-and-cancel, combine-a-fraction-and-cancel, rationalise-and-cancel — all converge on the same single formula, f′(x) = nxⁿ⁻¹, for the three shapes of rational exponent (positive integer, negative integer, fraction) that spec 4.2 says the rule must cover. That convergence is the actual mathematical content behind "the derivative as a limit" being a general definition rather than a trick specific to one function shape. No further sub-case (e.g. a general fractional n = p/q) is worked here: the algebra for that case is a direct extension of §2.3 with a qᵗʰ-root conjugate-style rationalisation, real but materially more intricate, and nothing in the facts bank or spec indicates WMA11 requires reproducing it — extending further would be adding depth the spec does not ask for, which is exactly the kind of padding the brief for this document rules out.

3. The second derivative: the same process, run on f′ instead of f

Spec 4.1 names f″(x) as permitted notation without requiring a separate definition — the lesson states, correctly, that f″(x) is "the derivative OF f′(x)... the exact same limiting-chord process... run a second time." Here is that claim executed, rather than only asserted, using §2.1's own cubic result so no new derivation has to be trusted on faith.

From §2.1, f(x) = x³ gives f′(x) = 3x². Apply the limit definition again, this time to g(x) = 3x² = f′(x):

f''(a) = lim_{h→0} [g(a+h) − g(a)] / h = lim_{h→0} 3[(a+h)² − a²] / h

Expand (a+h)² − a² = 2ah + h² (the same expansion §2.1 used for n = 2, scaled by the constant 3 out front):

= lim_{h→0} 3(2ah + h²)/h = lim_{h→0} 3(2a + h) [valid for h ≠ 0, after cancelling h]

Let h → 0: f″(x) = 6x. This matches the ordinary result of differentiating 3x² by the power rule directly, confirming the second derivative is not a new kind of object requiring a new definition — it is one more application of the same limit, taken on the function that the first application produced.

One boundary is worth stating precisely rather than assumed: the technique-hub formula bank (§1.4.1) already flags, as a genuine, checked gap, that no primary-source WMA11 question reviewed in the facts bank uses f″(x) as a maximum/minimum classification test (the second-derivative sign test) — the one place f″(x) appears in a verified real question, Jan 2019 Q6(b) (facts bank §4.4/"Lesson-audit log," re-fetched live from qualifications.pearson.com and a physicsandmathstutor.com mirror: f(x) = 2x^(5/2) − 40x + 8, x > 0; f″(x) = 5 gives 7.5x^(1/2) = 5 → x = 4/9), uses it as an ordinary algebraic quantity to solve for, not as a stationary-point classifier. Correction made in this pass: an earlier version of this line cited "Jan 2025 Q5(b)" for this sighting; the facts bank contains no such citation, and the formula bank's own line 109 carries the same error uncorrected. The real, independently re-checkable anchor is Jan 2019 Q6(b), confirmed above and cross-checked against the SIGNAL topic brief for this same spec point, which caught the same mistake independently. This document does not repeat the fetch's full detail, only its conclusion: second-order derivatives are confirmed-examinable as objects at WMA11, but their most common textbook use (classifying turning points) is not evidenced in this paper's own verified record.

4. The rate-of-change reading

Spec 4.1's guidance gives this reading in six words, verbatim: "dy/dx is the rate of change of y." It is the same number as the gradient reading, read through a different lens: if y = f(x) and dy/dx = 3 at some point, y is momentarily increasing three units for every one unit x increases, at that instant only — the value can be different at a different x, exactly because dy/dx is itself a function of x (spec 4.1's own point, restated from a different angle in §0 above). Beyond this one-sentence definition, the facts bank contains no WMA11 exam question that frames a differentiation result explicitly in rate-of-change language rather than gradient/tangent language — every verified differentiation question in §4.4 (Jan 2023 Q1, Jan 2025 Q5(a) and (c), Oct 2021 Q2, Oct 2021 Q6(c)) is phrased and marked in gradient/tangent terms. This is stated as an honest gap, not evidence that a rate-of-change framing could not appear — only that this pass found none.

5. Real exam-angle synthesis, and the one genuine finding this document contributes

The facts bank's §4.4 already documents, with full mark-scheme and examiner-report citations, five verified real WMA11 differentiation questions: Jan 2023 Q1 (differentiate then evaluate an exact gradient, with a confirmed "added +c after differentiating" trap), Jan 2025 Q5(a) (differentiating a negative-index term, 2/x → −2x⁻²), Oct 2021 Q2 (setting f′(x) equal to a given gradient value and solving the resulting equation), Oct 2021 Q6(c) (tangent at a stationary point, the zero-gradient trap), and Jan 2025 Q5(c) (tangent-line problem with a domain-restriction trap). Those citations, and the trap analysis built on them, already live in codex-formula-and-method-bank.md §1.4.1–1.4.3 and are not reproduced here.

What has not previously been stated explicitly, and is the actual additive claim of this section, is what those five questions have in common at the level of spec 4.1 specifically: every one of them starts from an f′(x) the student is expected to obtain by the power rule (spec 4.2) and then uses that f′(x) as an input to a further step — evaluating it, solving an equation built from it, or building a tangent/normal from it. None of the five requires writing out, or reasoning about, the limit lim_{h→0}[f(a+h)−f(a)]/h itself. Put in the facts bank's own evidentiary convention: a first-principles / limit-definition differentiation question is seen in 0 of 5 fully-mined WMA11 series in the facts bank (Jan 2019, Oct 2020, Oct 2021, Jan 2023, Jan 2025). Two further series (Oct 2022, Jan 2024) were confirmed as genuine matching documents but not mined question-by-question, and two more (June 2022, June 2023) were checked for one targeted question each unrelated to differentiation — so this is a statement about what the facts-bank pass actually found, not a claim that no WMA11 series in existence has ever asked a first-principles question.

The honest conclusion for a student using this document: at WMA11, spec 4.1's limit definition is, on the verified evidence available, tested through its consequences (the power rule it justifies, and what you do with the resulting f′(x)) rather than through being asked to reproduce the limit argument on demand. Understanding §2–§3 above is what makes the power rule in spec 4.2 something other than an arbitrary recipe — which is the spec's own stated reason for including 4.1 as a separate item from 4.2 in the first place — but the verified record gives no basis for predicting a standalone "differentiate from first principles" question on this specific paper.

6. Sources

Correction log. 2026-09-29 (same-day re-verification pass): §3's second-derivative citation corrected from a non-existent "Jan 2025 Q5(b)" to the real anchor, Jan 2019 Q6(b) (f(x) = 2x^(5/2) − 40x + 8, f″(x) = 5 ⟹ x = 4/9), re-confirmed by grep against WMA11-verified-facts.md. codex-formula-and-method-bank.md line 109 carries the same uncorrected error and is out of scope for this file to fix directly.