Additional Mathematics Synthesis Guide 10: Tangents, Normals, Stationary Points and Optimisation

Learning route: Understand the information · Build the lines · Classify the points · Optimise a model · Attempt the practice · Check independent control.

Scope: the 2026 O-Level Additional Mathematics syllabus covers tangents, normals, stationary points, connected rates and maxima/minima applications. Check the learner’s actual syllabus separately. These are original learning problems, not official or predicted examination questions.

1. The derivative is a local rate, not a replacement for the function

For y = f(x), the value f(a) tells us the height of the graph at x = a. The value f′(a) tells us its local gradient there, when the derivative exists. These are different quantities. Substituting a into f′ when the point’s y-coordinate is required is a category error, not a small arithmetic slip.

For f(x) = x², the gradient between x = a and x = a + h is ((a + h)² − a²)/h = 2a + h for h ≠ 0. As h approaches zero, this secant gradient approaches 2a. Thus f′(a) = 2a. The derivative describes the limiting local direction rather than the gradient of a chord of some fixed width.

The same notation works beyond pictures. If C(x) is a cost measured in dollars and x is a number of units, C′(x) has units of dollars per unit. If s(t) is displacement in metres, s′(t) is velocity in metres per second. Naming the input, output and units helps preserve meaning through the algebra.

A useful habit is to label an intermediate result before using it: “f′(2) = 9, so the tangent gradient is 9.” The next step then has a clear mathematical role.

2. Read the expression before choosing a differentiation rule

Not every expression should be expanded. In (2x − 1)⁴, the outer operation is a fourth power and the inner function is 2x − 1. The chain rule gives 8(2x − 1)³ directly. Expansion is possible, but it creates several more coefficients to control without helping the derivative.

In x²e^−x, the expression is a product. Differentiation gives 2xe^−x − x²e^−x = xe^−x(2 − x). The factored form is particularly useful for finding where the derivative is zero and for determining its sign. Expanding or rounding would hide exactly the structure needed next.

For y = (x + 1)/(x − 2), the quotient rule gives y′ = [(x − 2) − (x + 1)]/(x − 2)² = −3/(x − 2)², with x ≠ 2. The negative derivative tells us the function decreases on each of the intervals separated by x = 2. It does not justify describing the function as decreasing across a gap where it is undefined.

The standard trigonometric derivative formulas use radians. For example, d[sin(2x)]/dx = 2cos(2x) when x is in radians. A numerical evaluation must use the corresponding angle unit. The symbolic rule and the calculator state need to describe the same quantity.

3. A tangent equation needs both a point and a gradient

At x = a on a differentiable graph, the tangent passes through (a, f(a)) and has gradient f′(a). Therefore its equation is y − f(a) = f′(a)(x − a). This is ordinary coordinate geometry after calculus has supplied the local gradient.

Separate the two substitutions. Obtain the point from the original function and the gradient from the derivative. Then combine them in the point-gradient equation. Writing these on separate lines reduces the chance of placing the gradient in the point’s y-coordinate or using the curve’s value as the gradient.

The normal is the line through the same point perpendicular to the tangent. When the tangent has a finite, non-zero gradient m, the normal gradient is −1/m. If the tangent is horizontal, the normal is vertical instead. There is no finite negative reciprocal of zero.

4. Worked example: tangent and normal to a cubic

Find the tangent and normal to y = x³ − 3x + 2 at x = 2.

The point is (2, 4), because 2³ − 3(2) + 2 = 4. The derivative is 3x² − 3, so the tangent gradient is 9. The tangent is y − 4 = 9(x − 2), or y = 9x − 14.

The normal gradient is −1/9. Its equation is y − 4 = −(x − 2)/9, or x + 9y = 38. Both equations pass through the same point, and the finite gradients multiply to −1.

Check the tangent more deeply. Equating the line and curve gives x³ − 12x + 16 = 0, which factors as (x − 2)²(x + 4) = 0. The line is tangent at x = 2 but also meets the curve at x = −4. Tangency is local; it does not mean that the line is forbidden from meeting a general curve somewhere else.

This is a useful correction to an overgeneralised picture. For a line and a non-degenerate quadratic graph, the repeated-root method describes tangency cleanly. For more general curves, “exactly one intersection” is not a universal definition of tangent.

5. A tangent can cross the curve at the point of contact

For y = x³ at x = 0, the derivative is zero, so the tangent is y = 0. Yet x³ is negative to the left of zero and positive to the right. The curve crosses its tangent at the point of contact.

Conversely, the line y = 1 has only one intersection with y = x³, at x = 1, but it is not tangent there. The curve’s derivative at 1 is 3, whereas the line’s gradient is zero. Counting intersections alone has failed as a test.

These examples matter because they distinguish a definition from a familiar illustration. The derivative supplies the correct local direction. A sketch of a line “just touching” a parabola is useful, but it should not become a rule that excludes legitimate tangents to other curves.

6. Worked example: the horizontal-tangent exception

For y = x² − 4x + 7, find the tangent and normal at its stationary point. Differentiation gives y′ = 2x − 4, so the stationary x-coordinate is 2. The point on the curve is (2, 3).

The tangent is horizontal: y = 3. The normal is vertical: x = 2. Writing “normal gradient = −1/0” does not produce a usable gradient and should not be treated as an ordinary numerical result.

The geometric requirement of perpendicularity remains valid. Only the finite-gradient representation needs to change. This is the same principle encountered when a vertical line cannot be expressed in the form y = mx + c with finite m.

7. A stationary point is a candidate, not a classification

A stationary point on a differentiable graph has f′(a) = 0. It may be a local maximum, a local minimum or a stationary point of inflexion. Solving f′(x) = 0 locates candidates but does not tell us which kind each one is.

The first-derivative test examines signs on either side. Positive to negative means the function changes from increasing to decreasing, giving a local maximum. Negative to positive gives a local minimum. If the derivative keeps the same sign, the point is not a local turning maximum or minimum.

At a stationary point, f″(a) > 0 gives a local minimum and f″(a) < 0 gives a local maximum under the standard smoothness conditions. If f″(a) = 0, that test is inconclusive. The discussion in OpenStax, Derivatives and the Shape of a Graph sets out these distinctions.

Do not replace “inconclusive” with “point of inflexion”. A test that supplies no conclusion has not established a different conclusion.

8. Worked example: compare two classification methods

Let f(x) = x³ − 6x² + 9x + 1. Then f′(x) = 3(x − 1)(x − 3). The stationary values of x are 1 and 3, giving the points (1, 5) and (3, 1).

For x < 1, both factors are negative, so f′ is positive. For 1 < x < 3, the factors have opposite signs, so f′ is negative. For x > 3, both are positive, so f′ is positive. Thus (1, 5) is a local maximum and (3, 1) is a local minimum.

The second derivative gives the same result more quickly: f″(x) = 6x − 12. At x = 1, f″ = −6; at x = 3, f″ = 6. This is an efficient classification when the second derivative is non-zero, while the sign chart explains the actual increase/decrease behaviour.

Keep the full coordinates in the conclusion. The numbers 1 and 3 answer where the derivative is zero; they are not the complete stationary points. The original function supplies the y-coordinates.

9. Three functions with the same inconclusive test result

At x = 0, each of f(x) = x⁴, g(x) = −x⁴ and h(x) = x³ has first derivative zero and second derivative zero. Their behaviours are nevertheless different.

x⁴ is non-negative and equals zero at the origin, so it has a minimum there. −x⁴ is non-positive and equals zero there, so it has a maximum. x³ is increasing through the origin and changes concavity there, so it has a stationary point of inflexion rather than a maximum or minimum.

The first derivatives confirm the distinctions. For x⁴, 4x³ changes from negative to positive. For −x⁴, −4x³ changes from positive to negative. For x³, 3x² remains positive on both sides of zero. A short comparison like this is a stronger repair than memorising one more isolated “exception”.

10. Local and global questions require different checks

Return to f(x) = x³ − 6x² + 9x + 1, but now ask for its greatest and least values on 0 ≤ x ≤ 5. The local maximum at x = 1 is only one candidate for the greatest value on the whole interval.

Evaluate the stationary candidates and endpoints: f(0) = 1, f(1) = 5, f(3) = 1 and f(5) = 21. The greatest value is 21 at x = 5. The least value is 1 at x = 0 and x = 3. The local maximum of 5 is not the global maximum on this interval.

For continuous functions on closed bounded intervals, the standard method compares all interior critical candidates with the endpoints. The modelling treatment in OpenStax, Applied Optimization Problems explains this approach. If a function is not differentiable at an interior point, that point may also need examination.

An open interval is different. For f(x) = x on 0 < x < 1, neither endpoint is allowed, and there is no attained largest or smallest value. Stating the domain is therefore part of answering the optimisation problem, not an optional note after the calculation.

11. Optimisation begins before differentiation

An optimisation problem needs an objective and constraints. The objective is the quantity to be made as large or small as possible. The constraints describe the permitted configurations. The derivative becomes useful after those have been expressed as a function of a suitable variable.

A common failure is to differentiate whichever formula is most familiar. In a rectangle problem, the student may differentiate the perimeter even though the question asks for maximum area. Another may obtain a correct area formula in two variables and differentiate while silently holding fixed a variable that is actually linked to the first by a constraint.

Write the objective with its meaning: A(w) is area as a function of width; V(x) is volume after cutting squares of side x; D²(x) is squared distance from a moving point on a curve. Then use the constraint to eliminate any dependent variable. Finally state the domain that makes the configuration physically and mathematically valid.

12. Worked model: three sides of a rectangular enclosure

A rectangular enclosure uses a straight wall as one side. There are 48 m of fencing for the other three sides. Find the greatest possible area under this idealised model.

Let w be the distance out from the wall and L the length along it. The fence constraint is 2w + L = 48. Therefore L = 48 − 2w and A(w) = w(48 − 2w) = 48w − 2w², with 0 < w < 24.

Differentiation gives A′(w) = 48 − 4w, so the stationary candidate is w = 12. Then L = 24 and the area is 288 m². A″(w) = −4 confirms the local maximum.

Completing the square supplies a global check: A(w) = 288 − 2(w − 12)² ≤ 288, with equality at the allowed value w = 12. This verifies the maximum over the whole permitted domain, not merely in a small neighbourhood.

The model assumes a sufficiently long straight wall, negligible fence thickness and no gates or other constraints. Introducing a gate or a maximum permitted depth would create a different feasible set. The calculation should return to the situation it actually models rather than claim a universal design recommendation.

13. Worked model: cutting an open box from a sheet

A 30 cm by 20 cm rectangular sheet has a square of side x cut from each corner. The sides are folded up to make an open box. Find the cut size that maximises the idealised volume.

The height is x, the base length is 30 − 2x and the base width is 20 − 2x. Thus V(x) = x(30 − 2x)(20 − 2x) = 600x − 100x² + 4x³. All three dimensions must be positive, so 0 < x < 10.

The derivative is V′(x) = 600 − 200x + 12x². Setting it to zero gives 3x² − 50x + 150 = 0, hence x = (25 ± 5√7)/3. The two candidates are approximately 3.92375 and 12.7429.

The second candidate is outside the domain. It would make the base width negative and cannot represent the box. The valid candidate is x = (25 − 5√7)/3 cm. The derivative is positive before it and negative after it within the allowed interval, so volume rises and then falls.

The dimensions are approximately 22.1525 cm by 12.1525 cm by 3.92375 cm, giving volume approximately 1056.31 cm³. These extra digits show the numerical calculation; round the final reported answer to the accuracy the question requests.

What should the learner explain? The rejected root is not rejected because it looks inconvenient. It violates a stated geometric condition. Conversely, a stationary candidate inside the domain still needs classification. Solving the derivative equation is only one part of the reasoning.

14. Connected rates: differentiate the relationship before substituting an instant

Suppose a circle’s radius r changes with time and its area is A = πr². The area and radius are not independent time-varying quantities. Differentiating with respect to time gives dA/dt = 2πr × dr/dt.

If r = 5 cm at a particular instant and dr/dt = 0.2 cm/s, then dA/dt = 2π(5)(0.2) = 2π cm²/s. The unit is an area per time, not a length per time.

Substituting r = 5 into A = πr² before differentiating would turn the relationship into the numerical statement A = 25π for that instant. Differentiating that constant as though it described all times loses the variation we were trying to measure. The value at an instant and the function through time are different objects.

Reverse the problem: if dA/dt = 6π cm²/s when r = 3 cm, then dr/dt = 1 cm/s. If the area continues to grow at that same rate when r = 6 cm, the radius rate is 0.5 cm/s. A constant area rate does not imply a constant radius rate.

15. Connected investigation: the closest point and the normal

Extension investigation. Find the points on y = x² closest to Q = (0, 3). Then explain why the joining line is normal to the curve at each closest point.

A general point on the curve is P = (x, x²). Its squared distance from Q is D² = x² + (x² − 3)² = x⁴ − 5x² + 9. Minimising D² is equivalent to minimising D because the square-root function is increasing for non-negative inputs. The squared expression is easier to differentiate.

Let F(x) = D². Then F′(x) = 4x³ − 10x = 2x(2x² − 5). The stationary candidates are x = 0 and x = ±√(5/2). At x = 0, F″(0) = −10, so that candidate is a local maximum of squared distance, not the closest point.

At x = ±√(5/2), the points are (±√(5/2), 5/2). Completing the square in x² gives F(x) = (x² − 5/2)² + 11/4. Thus the global minimum squared distance is 11/4, and the minimum distance is √11/2.

Write either non-zero x-coordinate as a. The tangent gradient at P is 2a. The gradient of PQ is (3 − 5/2)/(0 − a) = −1/(2a). These gradients multiply to −1, so PQ is normal to the curve.

This investigation connects coordinate distance, an algebraic simplification, differentiation, classification and the geometric normal condition. Symmetry predicts the pair of equally good points. The completed square confirms the global result, while perpendicularity checks the geometry independently.

16. A negative derivative needs an interval

Return to f(x) = (x + 1)/(x − 2). Its derivative is negative wherever the function is defined. It is nevertheless misleading to say that every larger permitted x produces a smaller value without noting the discontinuity. For example, f(1) = −2 and f(3) = 4.

The function decreases on each connected interval (−∞, 2) and (2, ∞). The usual derivative-sign reasoning is applied on an interval where the required regularity conditions hold. The missing point at 2 separates the two pieces of the graph.

This is a useful example of why domain information should travel with the formula. A perfectly correct derivative can lead to an incorrect global statement when the interval has been forgotten.

17. Checking should target a different possible failure

For a tangent, check both point membership and gradient. For a normal, check the shared point and perpendicular direction. For a stationary point, check that the derivative is zero and that the classification follows from a valid test. For an optimum, check the domain, all relevant candidates and the requested quantity.

Different checks catch different errors. Substitution cannot by itself establish that a line is tangent. A negative second derivative at a candidate does not establish that it beats every endpoint. A plausible decimal volume does not show that the dimensions satisfy the original sheet constraint.

A calculator can help evaluate candidates, but it does not choose the feasible domain for the model. It also does not automatically distinguish an excluded root from an inconvenient one. Those remain mathematical decisions that should be visible in the working.

18. Practice: identify the requested object

Task 1. Find the tangent and normal to y = (3x + 1)⁵ at x = 0. Explain where the factor 3 enters the derivative.

Task 2. Find the tangent and normal to y = x³ − 3x at x = 1. Why is the usual normal-gradient expression not the right final form here?

Task 3. Find and classify every stationary point of y = x⁴ − 4x².

Task 4. Find the greatest and least values of y = x³ on −1 ≤ x ≤ 2. Explain the role of the stationary point at zero.

Task 5. Differentiate y = (x² + 1)/(x + 1), state its domain and locate the stationary x-values. Do not cancel x + 1 from separate terms.

Task 6. Repeat the three-sided enclosure model with 60 m of fencing. Find the dimensions and greatest area, and verify the result without differentiation.

Task 7. A circle’s radius grows at 0.4 cm/s. Find the area-growth rate when r = 10 cm. State the units.

Task 8. Find the point on y = 4/x, x > 0, closest to the origin. Check whether the line joining that point to the origin is normal to the curve.

Task 9. Explain why “a tangent has exactly one intersection with the curve” is not a correct definition for all differentiable curves. Give one example of a tangent crossing its curve and one example of a non-tangent having a single intersection.

19. Explained solutions

Task 1. The chain rule gives y′ = 5(3x + 1)⁴ × 3 = 15(3x + 1)⁴. At x = 0 the point is (0, 1) and the gradient is 15. The tangent is y = 15x + 1; the normal is x + 15y = 15. The factor 3 is the derivative of the inner function, not an optional multiplier.

Task 2. The derivative is 3x² − 3, which is zero at x = 1. The point is (1, −2). The tangent is y = −2 and the normal is x = 1. A vertical normal has no finite gradient, so writing −1/0 is not a substitute for the line equation.

Task 3. y′ = 4x(x² − 2), so x = 0 or x = ±√2. The second derivative is 12x² − 8. At zero it is −8, giving a local maximum at (0, 0). At ±√2 it is 16, giving local minima at (±√2, −4). The two minima have equal heights because the original function is even.

Task 4. The derivative 3x² is positive except at zero, and the function increases through zero. Compare the endpoints: y(−1) = −1 and y(2) = 8. The least value is −1 and the greatest is 8. The stationary point is not a turning maximum or minimum and does not replace the endpoint comparison.

Task 5. The derivative is (x² + 2x − 1)/(x + 1)², with x ≠ −1. Setting the numerator to zero gives x = −1 ± √2; both are in the domain. The denominator restriction must be retained even though the derivative has a convenient numerator equation.

Task 6. With width w, length L = 60 − 2w and area A = 60w − 2w². Differentiation gives w = 15, hence L = 30 m and area 450 m². The independent check is A = 450 − 2(w − 15)² ≤ 450. Equality occurs at an allowed positive width and length.

Task 7. dA/dt = 2πr dr/dt = 2π(10)(0.4) = 8π cm²/s. The changing quantity requested is area, so the units must include square centimetres rather than centimetres.

Task 8. Squared distance is F(x) = x² + 16/x². Differentiation gives F′ = 2x − 32/x³, so x⁴ = 16. With x > 0, x = 2 and y = 2. The identity F(x) = (x − 4/x)² + 8 proves the global minimum distance is 2√2. The curve gradient at x = 2 is −4/2² = −1, while the line from the origin to (2, 2) has gradient 1. They are perpendicular.

Task 9. The tangent y = 0 to y = x³ at the origin is crossed by the curve. The line y = 1 meets that same cubic only at x = 1 but is not tangent, because its gradient 0 differs from the curve gradient 3 there. The correct test concerns the local tangent direction, not a global count of intersections.

20. A teaching sequence that reveals the actual difficulty

Begin with one function and change the request: find its value, its gradient, a tangent, a normal, a stationary point and its greatest value on a restricted interval. This keeps much of the calculation familiar while changing the mathematical job. The learner must decide what each intermediate expression means.

If the derivative itself is unreliable, repair the relevant algebra or rule selection. If the derivative is correct but the line is wrong, inspect the point-gradient conversion. If a maximum is misidentified, inspect classification and domain. These are narrower and more teachable questions than “the student is weak in calculus”.

Then change the representation. Give a sign description of f′ rather than a formula for f. Ask the learner to describe increase, decrease and possible turning points without integrating. Or give a geometric constraint and ask only for the one-variable objective and domain. This isolates the decision that a full worked problem might otherwise conceal.

For an independent return task, remove the chapter label and ask the learner to explain why differentiation belongs. A response such as “I need the minimum distance, so I will minimise squared distance over the allowed curve points” shows a more precise decision than “this looks like calculus”. The actual calculation still needs to be checked; an excellent plan does not guarantee correct execution.

21. Continue from local change to accumulation

Differentiation describes local change. Integration answers a different family of questions about accumulation, area and net movement. The two are connected, but neither should be chosen merely because the other appeared earlier in the question.

Continue within this batch: Guide 9: Plane Geometry, Coordinate Geometry, Trigonometry and Proof · Guide 11: Definite Integrals, Area, Accumulation and Motion · Guide 12: Mixed-Topic Recognition, Method Selection, Verification and Transfer.

Return to the learning map: Additional Mathematics Directory · BTT Mathematics Hub. For the earlier repeated-root treatment of line–quadratic tangency, use Synthesis Guide 1.

Sources and boundaries: the linked SEAB syllabus supplies the stated 2026 scope; the two OpenStax sections support the standard derivative tests and constrained-optimisation method. The worked examples, counterexamples, investigations and practice solutions are original. The closest-point investigation is an optional synthesis extension, not a claim about a compulsory question type.