Did you know? A tangent line can turn a rough guess for √2 into a much better answer in just one calculation. Starting with x=1, a particular Newton step gives 1.5; the next gives 17/12, or about 1.4167. Yet the same technique, used on a different function, can cycle forever between two incorrect guesses. Understanding both outcomes is part of mathematical maturity.
For JC2 students and parents exploring a JC2 H3 Mathematics tutor in Bukit Timah, Newton–Raphson root finding is an excellent setting for H2 calculus, mathematical modelling, limits and rigorous explanation. Its numerical method is linked to the familiar MF27 formulae and H3 problem-solving expectations; it is presented here as a connected application, not as a separately named additional H3 chapter.
Quick answer: Given differentiable f with a nonzero derivative at the current iterate, Newton’s method uses xₙ₊₁=xₙ−f(xₙ)/f′(xₙ). Geometrically it uses the tangent line at (xₙ,f(xₙ)) to predict a better x-intercept. Near a simple root with suitable regularity, it often converges rapidly. A poor starting value, zero derivative, multiple root or cycle can prevent the expected convergence.
The best preparation combines exploration with verification. A calculator may calculate iterations, but a proof of convergence must show why they remain in an appropriate region and approach the intended root. The official H3 syllabus encourages logical reasoning and permits approved graphing calculators within stated assessment conditions.
The tangent-line derivation
Take a differentiable function f and a current approximation xₙ. Near that point, the tangent approximation is f(x)≈f(xₙ)+f′(xₙ)(x−xₙ). The method finds where this tangent line reaches height zero.
Set the approximate value to zero. Provided f′(xₙ)≠0, rearrange to obtain x=xₙ−f(xₙ)/f′(xₙ). Define that x as the next guess xₙ₊₁. Every term in the formula has a purpose: f(xₙ) is the vertical error and f′(xₙ) controls how much horizontal adjustment the tangent suggests.
The tangent line may be a good local approximation near a smooth simple root, but it is not the original curve. Its intercept can be far away if the derivative is nearly zero or the initial guess lies in an unsuitable region.
Do not claim the iteration is defined at every real x. Division by f′(xₙ) requires a nonzero derivative. If the derivative vanishes, another starting point or method may be needed. A calculator can also encounter numerical instability when the derivative is extremely small.
The goal of a root-finding problem should be stated: one approximate root, all real roots, an error bound or a proof of convergence. Newton’s method can help with the first, but the others require additional work.
For a function f(x)=x²−2, the derivative is 2x, so Newton’s method gives xₙ₊₁=xₙ−(xₙ²−2)/(2xₙ), which simplifies to (xₙ+2/xₙ)/2 for nonzero xₙ. The latter is sometimes easier to evaluate accurately.
Full worked example: approximating √2
Let f(x)=x²−2, and start with x₀=1. The first update is (1+2/1)/2=3/2=1.5. The second update is ((3/2)+2/(3/2))/2=(3/2+4/3)/2=17/12≈1.4166667.
The third update is ((17/12)+2/(17/12))/2=(17/12+24/17)/2=577/408≈1.4142157. The actual positive root √2 is about 1.41421356, so the third rational estimate is already close.
Numerical closeness alone does not prove the iteration converges from every starting point. For a full proof when x₀>0, first note that xₙ₊₁ is the arithmetic mean of two positive numbers xₙ and 2/xₙ.
By AM–GM, xₙ₊₁≥√[xₙ(2/xₙ)]=√2. Thus after one update, every subsequent iterate is at least √2. Also xₙ₊₁−xₙ=(2−xₙ²)/(2xₙ), which is nonpositive when xₙ≥√2.
Therefore from x₁ onward the sequence is nonincreasing and bounded below by √2. It converges to a limit L≥√2. Passing to the continuous update formula gives L=(L+2/L)/2, so L²=2, and the positive limit is L=√2.
This is a genuine convergence proof. It establishes positivity, an invariant lower bound, monotonicity and the final limit. The fixed-point equation L²=2 is only the last step; without convergence, it would not establish a limit.
The error has an exact identity: if r=√2 and xₙ>0, then xₙ₊₁−r=(xₙ−r)²/(2xₙ). The error is squared and divided by a positive factor. Once estimates are close to r, this explains the rapid improvement.
For a negative starting x₀, the same formula may lead to the negative root −√2. Thus the starting sign matters if the question specifically seeks the positive solution.
A general example: one real root of x³−x−1
Consider f(x)=x³−x−1, which is continuous. At x=1, f(1)=−1; at x=2, f(2)=5. The intermediate value theorem guarantees at least one root between one and two.
The derivative is f′(x)=3x²−1. Starting with x₀=1 gives x₁=1−(−1)/2=1.5. At x₁=1.5, f=0.875 and f′=5.75, so x₂=1.5−0.875/5.75=31/23≈1.347826.
Further Newton steps approach approximately 1.324718. The decimal is useful, but it should be labelled an approximation. The sign-change argument provides existence; appropriate monotonicity or a separate argument is needed for uniqueness.
For x≥1, f′(x)=3x²−1≥2>0, so the function is strictly increasing there. Together with the sign change, this establishes exactly one root in the interval [1,2]. That does not by itself address the possibility of a root elsewhere on the real line.
A full real-root analysis can examine the stationary points at x=±1/√3. The cubic’s local extrema are both negative, so it has just one real root. A tutor can choose the amount of analysis appropriate to the particular question.
The important distinction is that a numerical procedure estimates a root, while continuity and derivative reasoning establish what the estimated number represents and whether it is unique.
A striking counterexample: Newton can cycle
Let f(x)=x³−2x+2 and start Newton’s method at x₀=0. Here f(0)=2 and f′(0)=−2, so the next value is x₁=0−2/(−2)=1.
At x=1, f(1)=1 and f′(1)=1, so x₂=1−1/1=0. The calculation has returned to its starting point. It will now repeat 0,1,0,1,… indefinitely unless something changes.
This is a rigorous counterexample to the claim that Newton’s method always converges. Both updates are defined, and the cycle can be proved exactly without relying on a graph.
Nevertheless the function has a real root. For example f(−2)=−2 and f(−1)=3, so continuity guarantees a root between −2 and −1. The failure comes from the initial guesses and the tangent iteration, not from an absence of solutions.
A good mathematical response separates these statements: a root exists; this particular Newton sequence does not converge to it; a different initial value or a bracketing method may succeed.
An H3 learner should be encouraged to search for counterexamples to universal claims about algorithms, just as they would for an inequality or number-theory theorem.
Zero derivative, multiple roots and numerical caution
The Newton formula contains f′(xₙ) in its denominator. For f(x)=x²−2 and x₀=0, the derivative is zero, so a Newton update is undefined. If a numerical tool displays an error, the mathematics explains why.
Consider f(x)=x²+1. It has no real zero. Starting at x₀=1 gives x₁=1−2/2=0, after which f′(0)=0. This illustrates both the absence of a real root and a breakdown of the proposed iteration.
Multiple roots can change the convergence speed. For f(x)=(x−1)² and x≠1, Newton simplifies to xₙ₊₁=(xₙ+1)/2. The error xₙ−1 is halved, giving linear rather than the usual rapid quadratic convergence at a simple root.
At the root x=1, the derivative is zero and the raw formula has zero divided by zero. The correct action is to stop because the function value is already zero, not to perform an undefined update.
Near a nearly horizontal tangent, the ratio f/f′ can become very large. A candidate may jump outside an interval in which the desired root was initially bracketed. A robust numerical strategy may check whether the proposed step remains sensible and otherwise use a safer alternative.
Do not equate an apparently stable decimal display with a proved error bound. Limited display precision can hide changes, and a small difference between successive iterates need not always mean the function residual is small.
Bisection is slower but has a useful guarantee
Suppose f is continuous on [a,b] and f(a)f(b)<0. Then a sign change guarantees at least one zero inside. Bisection evaluates the midpoint, selects a half-interval retaining a sign change and repeats.
After n bisections, the interval width is (b−a)/2^n. If the midpoint is reported as the approximation, the error relative to a root inside the maintained bracket is at most half the final interval width.
This guarantee differs from Newton’s method. Bisection relies on a sign-changing bracket and continuity; it does not need derivatives. Newton may converge much faster near a suitable simple root but can cycle or become undefined.
For f(x)=x²−2 on [1,2], the initial values have opposite signs. The first midpoint 1.5 gives a positive function value, so the root lies in [1,1.5]. The next midpoint 1.25 is below √2, narrowing the bracket to [1.25,1.5].
The method produces an explicit interval containing a zero, making its error bound interpretable. If a question requires a certified numerical enclosure, bracketing can be more appropriate than an unqualified Newton display.
These are complementary tools. A learner should be able to choose based on the question’s target, available function information, starting values and required justification.
How to evaluate a numerical method responsibly
Before the first iteration, determine the function’s domain and whether the selected starting value is permitted. Check differentiability at that point and confirm the derivative is not zero.
During the iteration, record enough working to reconstruct the update when a question requires steps. Use consistent rounding and avoid introducing large rounding errors by truncating each intermediate value excessively.
After several iterations, evaluate the function at the candidate. A small residual |f(x)| is informative, but it does not always guarantee a small root error if the derivative is very small. In a region where |f′(x)| is bounded below, the mean value theorem can connect residual size to distance from a root.
If a bracketing interval is available, preserve it or periodically check the candidate against it. A sketch can reveal additional roots or problematic stationary points but does not replace a proof that all required roots have been accounted for.
For an exact theoretical claim about convergence, identify the invariant interval, monotonicity or local contraction conditions used. A fixed point of the Newton update is not enough on its own to show every starting sequence converges.
The official 2027 H3 syllabus expects access to an approved graphing calculator without computer algebra and generally permits unsupported calculator answers unless a question states otherwise. Where mathematical working is required, show mathematical steps rather than calculator commands, and sketch graphs used to find solutions.
This article supports those examination habits without suggesting that a particular Newton question must appear. It is an original connected application of H2 calculus, limits and the H3 emphasis on proof and non-routine reasoning.
The immutable eduKateSG small-group Mathematics reference near Sixth Avenue MRT documents a three-student Secondary 1 format, not a verified JC2 H3 class. Enquire directly about current advanced Mathematics tutoring availability.
60 original root-finding and convergence clinics
1. Tangent derivation
Question: Where does xₙ₊₁=xₙ−f(xₙ)/f′(xₙ) come from?
Worked reasoning: Set the tangent approximation f(xₙ)+f′(xₙ)(x−xₙ) equal to zero and solve for x.
Learning check: The formula requires a nonzero derivative at the current iterate.
2. A linear function
Question: Apply Newton to f(x)=x−5 from any starting number.
Worked reasoning: Since f′=1, the next point is x−(x−5)=5, the exact root.
Learning check: For a nonconstant linear function Newton reaches the root in one step.
3. Another linear function
Question: Apply Newton to f(x)=2x+3.
Worked reasoning: The update is x−(2x+3)/2=−3/2, regardless of the starting value.
Learning check: A constant nonzero slope makes the tangent line identical to the function.
4. Square-root update
Question: Derive Newton’s iteration for f(x)=x²−2.
Worked reasoning: With f′=2x, the update simplifies to (x+2/x)/2 for x≠0.
Learning check: The derivative restriction remains necessary after algebraic simplification.
5. Square root from one
Question: What is the first update when x₀=1 for x²−2?
Worked reasoning: x₁=(1+2)/2=3/2=1.5.
Learning check: The first iterate can be an overshoot and still support convergence.
6. Second √2 update
Question: Calculate the next update from x₁=3/2.
Worked reasoning: x₂=(3/2+4/3)/2=17/12≈1.4166667.
Learning check: Exact fractions make numerical errors easier to trace.
7. Third √2 update
Question: Calculate the next update from 17/12.
Worked reasoning: x₃=(17/12+24/17)/2=577/408≈1.4142157.
Learning check: Rapid improvement follows from the square-error identity near the root.
8. Positive starting values
Question: Why does Newton for x²−2 remain positive if x₀>0?
Worked reasoning: Each update is the arithmetic mean of x and 2/x, both positive.
Learning check: An invariant domain keeps division valid during the iteration.
9. AM–GM lower bound
Question: Why is every positive update for x²−2 at least √2?
Worked reasoning: By AM–GM, (x+2/x)/2≥√[x·(2/x)]=√2.
Learning check: The inequality establishes a global lower bound after one update.
10. Decreasing after first update
Question: Why does the positive square-root iteration decrease once x≥√2?
Worked reasoning: The change equals (2−x²)/(2x)≤0 in that domain.
Learning check: A bound and a difference sign give monotonicity.
11. Complete convergence
Question: What proves convergence to √2 from x₀>0?
Worked reasoning: After the first step the sequence is bounded below by √2 and nonincreasing; its limit solves L²=2 with L>0.
Learning check: A fixed-point equation alone does not prove convergence.
12. Exact error formula
Question: How does the √2 error change?
Worked reasoning: xₙ₊₁−√2=(xₙ−√2)²/(2xₙ) for xₙ>0.
Learning check: Near a positive simple root, squaring small errors explains rapid local improvement.
13. A negative start
Question: What happens for f=x²−2 and x₀=−2?
Worked reasoning: The update stays negative and can converge to −√2.
Learning check: A method finding a root does not automatically select the desired sign.
14. Zero derivative at start
Question: Can Newton for x²−2 start at x₀=0?
Worked reasoning: No. The derivative 2x is zero, so the Newton fraction is undefined.
Learning check: The root-finding rule requires a valid tangent slope.
15. Another square root
Question: For f=x²−9, start at x₀=4. What is x₁?
Worked reasoning: x₁=(4+9/4)/2=25/8=3.125.
Learning check: The same method approximates any positive square root from a positive start.
16. Next √9 step
Question: From x₁=25/8 for x²−9, calculate x₂.
Worked reasoning: x₂=(25/8+72/25)/2=1201/400=3.0025.
Learning check: Verify a decimal against the original equation when checking precision.
17. Square root of four
Question: For f=x²−4 and x₀=3, find x₁.
Worked reasoning: x₁=(3+4/3)/2=13/6≈2.1667.
Learning check: Algebraic simplification reduces the chance of wrong derivative substitution.
18. Another √4 update
Question: From x₁=13/6, find the next value.
Worked reasoning: x₂=(13/6+24/13)/2=313/156≈2.00641.
Learning check: Successive exact fractions can be used as both computation and error checks.
19. Square root of three
Question: For f=x²−3 and x₀=2, what is x₁?
Worked reasoning: x₁=(2+3/2)/2=7/4=1.75.
Learning check: A suitable initial value makes the tangent method easy to calculate by hand.
20. Next √3 update
Question: From x=7/4 for x²−3, what is the next approximation?
Worked reasoning: x₂=(7/4+12/7)/2=97/56≈1.732142857.
Learning check: The estimate is close to √3≈1.732050808.
21. Cube root of two
Question: For f=x³−2 starting at 1.5, find the next Newton value.
Worked reasoning: The update is x−(x³−2)/(3x²), giving x₁=35/27≈1.296296.
Learning check: For a cubic, use the derivative 3x², not 2x.
22. Cube root second value
Question: Continue the cube-root iteration from 35/27.
Worked reasoning: The next approximation is about 1.260932, close to ∛2≈1.259921.
Learning check: A calculator may support arithmetic while the derivative formula supplies the method.
23. Cube root third value
Question: What happens after another Newton step?
Worked reasoning: The approximation is about 1.25992186, extremely close to the real cube root of two.
Learning check: Local convergence can be rapid for a suitable simple root and starting value.
24. Cubic sign bracket
Question: For f=x³−x−1, check f(1) and f(2).
Worked reasoning: f(1)=−1 and f(2)=5, so continuity guarantees a root in (1,2).
Learning check: A bracket proves existence separately from an iterative estimate.
25. Cubic first Newton update
Question: For x³−x−1, start at one.
Worked reasoning: With f(1)=−1 and f′(1)=2, x₁=1.5.
Learning check: Write both f and f′ values before simplifying the update.
26. Cubic second update
Question: From x=1.5 for x³−x−1, compute one more step.
Worked reasoning: f=0.875 and f′=5.75, so x₂=31/23≈1.347826.
Learning check: The tangent approximation corrects the initial overshoot.
27. Cubic third update
Question: What is the next approximation for that cubic?
Worked reasoning: x₃≈1.3252004, moving towards the root around 1.324718.
Learning check: An approximate numerical sequence is not a universal convergence theorem.
28. Monotone bracket for cubic
Question: Why is x³−x−1 strictly increasing on [1,2]?
Worked reasoning: Its derivative 3x²−1 is at least two there.
Learning check: Monotonicity plus a sign change proves one and only one root in that interval.
29. An oscillating Newton sequence
Question: For f=x³−2x+2 and x₀=0, compute x₁.
Worked reasoning: f(0)=2 and f′(0)=−2, so x₁=1.
Learning check: An initial point can produce a step away from the actual real root.
30. The second cycle step
Question: For the same cubic at x=1, compute the next iterate.
Worked reasoning: f(1)=1 and f′(1)=1, so x₂=0.
Learning check: Exact return to the start proves a two-cycle.
31. Cycle is not convergence
Question: Does the 0,1,0,1 Newton sequence converge?
Worked reasoning: No. Its even terms are zero and odd terms are one.
Learning check: A repeated two-cycle disproves the claim that Newton always converges.
32. A real root still exists
Question: Does f=x³−2x+2 have a real root despite the cycle?
Worked reasoning: Yes. f(−2)=−2 and f(−1)=3, so continuity gives a root between −2 and −1.
Learning check: Method failure is not the same as absence of a root.
33. No real zeros
Question: For f=x²+1, does a real root exist?
Worked reasoning: No. x²≥0 for all real x, so x²+1≥1.
Learning check: Check existence before trusting an iterative procedure.
34. Undefined second update
Question: For f=x²+1 and x₀=1, what happens?
Worked reasoning: The first update is zero; the derivative then vanishes and the next update is undefined.
Learning check: Even a well-defined first step may lead to a numerical breakdown.
35. A double root
Question: For f=(x−1)² and x≠1, simplify the Newton update.
Worked reasoning: It becomes xₙ₊₁=(xₙ+1)/2.
Learning check: A multiple root can have slower convergence than a simple root.
36. Double-root error
Question: How fast does the error shrink in that quadratic-square example?
Worked reasoning: xₙ₊₁−1=(xₙ−1)/2, so the error halves.
Learning check: This is linear convergence, not the usual quadratic local rate.
37. At the exact double root
Question: Can the raw Newton fraction be evaluated at x=1 for (x−1)²?
Worked reasoning: No. Both numerator and denominator vanish; the correct response is that the root has already been reached.
Learning check: Stopping criteria can avoid performing an undefined update.
38. A cosine equation
Question: Consider f=cos x−x with x₀=0.5. What is the first update?
Worked reasoning: The next value is about 0.755222, using f′=−sin x−1.
Learning check: Differentiate the whole function, including the negative x term.
39. Cosine second update
Question: What is the next value after about 0.755222?
Worked reasoning: Approximately 0.739142, approaching the root around 0.739085.
Learning check: Newton can work for transcendental equations as well as polynomials.
40. A small derivative
Question: Why can Newton take an excessively large step near f′(x)=0?
Worked reasoning: The correction f(x)/f′(x) divides by a tiny number, potentially producing a very large magnitude.
Learning check: A locally flat tangent may be a poor root predictor.
41. A root residual
Question: What does a small |f(xₙ)| tell us?
Worked reasoning: The candidate nearly satisfies the equation numerically, but that alone may not bound distance to a root without slope information.
Learning check: Residual and root error are distinct quantities.
42. Mean value bound
Question: If |f′(x)|≥m>0 throughout an interval joining a candidate and a root, what follows?
Worked reasoning: By the mean value theorem, the distance to that root is at most |f(candidate)|/m.
Learning check: A derivative lower bound can convert a residual into an error estimate.
43. Bisection prerequisites
Question: When is the basic sign-change bisection theorem applicable?
Worked reasoning: When f is continuous on [a,b] and f(a),f(b) have opposite signs.
Learning check: A sign-changing bracket guarantees a zero, unlike an arbitrary Newton start.
44. Bisection first step
Question: For x²−2 on [1,2], evaluate midpoint 1.5.
Worked reasoning: f(1.5)=0.25>0, so a root remains in [1,1.5].
Learning check: Use the sign to preserve a valid bracket.
45. Bisection second step
Question: For the same bracket [1,1.5], evaluate midpoint 1.25.
Worked reasoning: f(1.25)=−0.4375, so the root lies in [1.25,1.5].
Learning check: Every step halves the bracket while retaining a root.
46. Bisection width
Question: After n bisections of an interval of width W, what is the width?
Worked reasoning: W/2^n. The bracket keeps at least one root under the continuity and sign-change hypotheses.
Learning check: The interval length is an explicit numerical accuracy guarantee.
47. Midpoint error bound
Question: If a root is inside a bracket of width W, what is the midpoint error bound?
Worked reasoning: At most W/2, because no point inside is farther than half the bracket width from its midpoint.
Learning check: A certified enclosure is stronger than a displayed decimal alone.
48. Newton versus bisection
Question: Which method generally has a stronger bracket guarantee?
Worked reasoning: Bisection, provided a continuous sign-changing interval is maintained; Newton can be faster locally but need not stay in a bracket.
Learning check: Choose the algorithm according to the requested reliability and available information.
49. The secant idea
Question: What distinguishes the secant method from Newton’s method?
Worked reasoning: It estimates slope from two recent function values rather than evaluating the derivative analytically.
Learning check: A derivative-free method needs two starting points and can also have failure cases.
50. A proof of convergence
Question: Why is solving L=g(L) insufficient to show iterations converge?
Worked reasoning: The oscillating Newton example has an iteration that does not approach a limit, despite the function having a real root.
Learning check: First establish existence of the limit, then identify it.
51. Domain restriction
Question: Can Newton evaluate ln x at a negative iterate if working over the real numbers?
Worked reasoning: No. The logarithm is not real-valued there, so that iterate is outside the function’s domain.
Learning check: An update formula does not override original domain conditions.
52. Rounding discipline
Question: Why should intermediate values be kept at adequate precision?
Worked reasoning: Rounding at every step can accumulate numerical error or alter a later derivative evaluation.
Learning check: Report requested final accuracy while retaining sufficient internal precision.
53. Graphing window caution
Question: Can a narrow graphing-calculator window miss a root?
Worked reasoning: Yes. A root outside the selected interval is invisible in that view.
Learning check: A numerical sketch is a discovery aid, not proof of global completeness.
54. Graph-based response
Question: What should a candidate do if graphs are used to find a solution?
Worked reasoning: The official syllabus says candidates should sketch the graphs as part of their answer.
Learning check: Follow the question’s mathematical communication requirements.
55. Stopping at small steps
Question: Does |xₙ₊₁−xₙ| small always prove near-zero residual?
Worked reasoning: Not necessarily. Flat derivatives, rounding and unsuitable iteration behaviour can mislead; evaluate f(xₙ) too.
Learning check: Use more than one diagnostic when assessing a numerical answer.
56. Exact versus approximate
Question: How should √2 be reported if the question asks for an exact value?
Worked reasoning: Use √2 rather than a rounded Newton decimal.
Learning check: The form of the result must match the requested output.
57. A quadratic error identity
Question: For xₙ>0 in the √2 iteration, what exact error equation holds?
Worked reasoning: xₙ₊₁−√2=(xₙ−√2)²/(2xₙ).
Learning check: An algebraic identity can explain local speed more rigorously than observed decimals.
58. Positive convergence proof
Question: List the three ingredients for convergence of the positive √2 iteration.
Worked reasoning: Positivity, the AM–GM lower bound √2 and monotonic decrease after the first update.
Learning check: The monotone convergence theorem then supplies the limit.
59. Always converge?
Question: Does Newton converge from every allowed starting value on every smooth function?
Worked reasoning: No. The cubic x³−2x+2 cycles between zero and one from start zero.
Learning check: A single valid counterexample refutes a universal algorithm claim.
60. A mathematical justification
Question: What belongs in a proof-based numerical answer?
Worked reasoning: The function, derivative, starting value, valid update, requested iteration values and any required error or convergence argument.
Learning check: A calculator command list is not a substitute for mathematical notation when working is required.
Seven-session JC2 numerical-methods tutorial plan
Session one: derive Newton’s formula from the tangent line using a differentiable function. Practise differentiating simple polynomials and identify when the denominator would be zero. Ask why the tangent’s root is only an estimate of the curve’s root.
Session two: approximate √2 from one, maintaining exact fractions through three steps. Then prove positivity, AM–GM boundedness and monotonicity from the first update. Use the fixed-point equation only after convergence has been established.
Session three: compare x²−9 and x³−2. Calculate the first two iterates and check them by substituting back into the original functions, retaining sufficient precision until the final answer.
Session four: investigate x³−x−1 using continuity and derivative information. Separate proof of root existence, proof of uniqueness in a specified interval and the numerical approximation obtained from Newton’s method.
Session five: study failures. For x³−2x+2 starting at zero, derive the exact 0,1 cycle. Then consider f=x²+1 and a starting value leading to a zero derivative. Ask which assumption breaks in each example.
Session six: compare Newton and bisection on x²−2. Maintain a sign-changing bracket, compute widths and midpoint error bounds, then discuss when a fast tangent step might be preferable and when a guarantee matters more.
Session seven: solve an unfamiliar differentiable equation. The learner must choose an initial point, check the derivative, write the update, interpret the residual, identify any required graph sketch and state the limitation of the numerical answer.
For home practice, parents can ask why a tangent line estimates the root, whether the denominator is safe and what evidence shows the answer is accurate. This encourages mathematical explanation without needing to operate the calculator.
The eduKateSG immutable three-learner Mathematics reference describes Secondary 1 teaching near Sixth Avenue MRT, not a current H3 timetable. Enquire with eduKateSG about JC2 H3 support before arranging lessons.
Frequently asked questions
What is Newton–Raphson method?
A tangent-line iteration xₙ₊₁=xₙ−f(xₙ)/f′(xₙ), defined where the current derivative is nonzero, used to approximate roots.
Does Newton always converge?
No. A suitable simple root and sufficiently close starting value often give rapid convergence, but zero derivatives, cycles, unsuitable starts and multiple roots can cause problems.
Why is the derivative in the formula?
It represents the tangent-line slope. Dividing the current function value by that slope predicts how far to move horizontally toward the tangent’s x-intercept.
Can the first guess determine which root is reached?
Yes. Different starting values can lead to different roots, failure or cycling. A sign-changing bracket and derivative information can help select a safer strategy.
What is the difference between a residual and an error?
Residual is |f(x)| at the candidate; root error is the distance to an actual root. A slope bound or other analytical information is needed to connect them reliably.
Why can Newton converge slowly for a repeated root?
A repeated root has zero derivative at the root, and the usual simple-root quadratic convergence conclusion may fail. For f=(x−1)² the error is merely halved.
How is bisection different?
Bisection preserves a continuous sign-changing bracket and halves its width each step. It is generally slower but gives a reliable interval containing a root.
Is this a separately listed additional topic in H3 9820?
The 2027 syllabus assumes H2 calculus, includes limits, reasoning and approved graphing calculator use. Newton’s method here is a connected enrichment and numerical application, not a separately named additional H3 heading.
Can a calculator supply an unsupported answer?
The official 2027 H3 syllabus generally allows unsupported graphing-calculator answers unless a question says otherwise. When supporting work is required, show mathematical steps and notation.
How can a tutor check understanding?
Ask the learner to derive the tangent update, identify where it is undefined, compare one successful iteration with a failure case and justify the kind of answer requested.
Official 2027 syllabus and related learning
The 2027 H3 Mathematics (9820) syllabus prescribes proof and non-routine reasoning, with H2 calculus as assumed knowledge, and sets out graphing-calculator expectations. This Newton–Raphson guide develops those capabilities through original numerical applications.
Continue with recurrence relations and inductive bounds, MF27 and graphing-calculator checks and limit laws for connected JC2 problem-solving skills.