eduKate Singapore

VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Next useful routeCanonical hubHELP

Mathematics in Bukit Timah | Newton’s Method, Tangent Iteration and Root Convergence

—

by

eduKate Asia

Did you know that you can estimate √2 by repeatedly drawing a tangent to the curve y=x²−2? Start at x=1.5, draw the tangent at that point and see where the tangent crosses the horizontal axis. The crossing gives a better guess. Draw another tangent and the answer becomes more accurate still. This repeated process is Newton’s method, also called the Newton–Raphson method.

This Mathematics in Bukit Timah guide explains root-finding, tangent gradients, the Newton iteration formula, worked square-root calculations, residuals, convergence, failure cases, stopping conditions and comparison with bisection. Every graph, classroom measurement and example is fictional Mathematics—not a survey of Bukit Timah terrain, a school laboratory or a real equipment calibration. The point is to understand why the numerical procedure works, when it might fail and how to check the resulting answer.

Quick answer: To solve f(x)=0 with a differentiable function f, use xₙ₊₁=xₙ−f(xₙ)/f′(xₙ) whenever the derivative at the current estimate is nonzero. For f(x)=x²−2, this becomes xₙ₊₁=(xₙ+2/xₙ)/2 for xₙ≠0. From x₀=1.5, the first estimates are 1.4166667, 1.4142157, and approximately 1.4142136, approaching √2.

Article contents

Additional Mathematics learning material used for fictional Newton iteration, root convergence and tangent method examples.

What problem is Newton’s method solving?

A root or zero of f is an input r for which f(r)=0.

Some equations factor easily. For example x²−5x+6=0 factors into (x−2)(x−3)=0, giving exact roots two and three.

Other equations have roots that are irrational or cannot be expressed conveniently through elementary formulas. A numerical root-finding algorithm approximates a solution while keeping track of assumptions and error.

Why Bukit Timah is only a setting for the lesson

Families around Beauty World, Sixth Avenue and King Albert Park may picture a pupil sketching a curved graph or trying values in a calculator.

But our curve y=x²−2 is an abstract mathematical function. Its x-axis intersection is not a real hill crossing or property boundary.

The method can later be useful for physical models, yet any such application requires genuine measurements, a justified equation and an appropriate domain.

A root is not the same as a small function value

If f(x)=x²−2 and x=1.4, then f(1.4)=1.96−2=−0.04.

That is close to zero, but not zero. The input 1.4 is an approximate root, not the exact value √2.

A numerical algorithm provides candidate inputs. Substituting them back into the original equation is an important check.

Worked example 1: bracket root two

Let f(x)=x²−2.

At x=1, f(1)=−1. At x=2, f(2)=2. Since f is continuous, at least one root lies between one and two by the intermediate value theorem.

The positive root is √2. We can use the interval to make an informed initial guess for a numerical method.

Begin with a tangent line

Suppose a smooth curve y=f(x) passes through (xₙ,f(xₙ)).

Its tangent gradient at that point is f′(xₙ). The tangent line has equation y−f(xₙ)=f′(xₙ)(x−xₙ).

Newton’s idea is to replace the curve locally by that straight tangent, then use the tangent’s horizontal-axis crossing as the next guess.

Derive the Newton update from the tangent

Set y=0 in the tangent equation because we want the estimated root.

Then −f(xₙ)=f′(xₙ)(xₙ₊₁−xₙ).

Provided f′(xₙ)≠0, rearrange to obtain xₙ₊₁=xₙ−f(xₙ)/f′(xₙ). The update is derived from the tangent, not an arbitrary calculator trick.

A nonzero derivative is essential to the update

Newton’s formula divides by f′(xₙ). If the derivative at the current point is zero, that update is undefined.

A very small derivative can also produce an unusually large step, potentially moving the next guess far away from the intended root.

A mathematically valid root-finding plan checks slope and domain rather than assuming that every starting value is safe.

Worked example 2: obtain the iteration for square root two

Choose f(x)=x²−2, so f′(x)=2x.

Newton update is xₙ₊₁=xₙ−(xₙ²−2)/(2xₙ). Combine into one fraction or simplify:

xₙ₊₁=(xₙ+2/xₙ)/2, valid when xₙ≠0. This is the classical repeated averaging formula for √2.

First iteration from one and a half

Take x₀=1.5.

Compute x₁=(1.5+2/1.5)/2=(1.5+1.333333…)/2=1.416666666….

The original estimate was 0.0858 above √2; the new estimate is only about 0.00245 above it. The tangent step has improved accuracy considerably.

Worked example 3: keep an exact fraction

Write x₀=3/2. Then 2/x₀=4/3.

Average them: x₁=(3/2+4/3)/2=((9+8)/6)/2=17/12.

The fraction 17/12 represents the first iterate exactly. Its decimal 1.416666… is merely one numerical display of the same rational value.

Second iteration with fractions

Starting from x₁=17/12, use x₂=(17/12+2/(17/12))/2.

Since 2/(17/12)=24/17, the sum is 17/12+24/17=(289+288)/204=577/204.

Divide by two to obtain x₂=577/408≈1.414215686. The result is already very close to √2.

Worked example 4: verify the second estimate

Use x₂=577/408. Square it and compare with two.

The squared residual is x₂²−2=(577²−2×408²)/(408²)=(332929−332928)/166464=1/166464.

Thus the function value is positive and extremely small, approximately 0.0000060073. The estimate lies just above the positive root.

A third iteration gives more accuracy

Use x₂≈1.414215686 in x₃=(x₂+2/x₂)/2.

The next value is approximately 1.414213562. The first several decimal digits agree with √2.

The correct conceptual statement is that iteration approaches the root under suitable conditions, not that a finite decimal becomes literally equal to an irrational number.

Worked example 5: compare estimates by residual

For x₀=1.5, residual f(x₀)=0.25.

For x₁=17/12, f(x₁)=(289/144)−2=1/144≈0.006944.

For x₂=577/408, f(x₂)=1/166464≈0.000006007. The residual shrinks rapidly in this example.

Why Newton’s method is fast near a simple root

For a sufficiently smooth function whose root r has f′(r)≠0, Newton’s method often has quadratic local convergence.

Informally, once the approximation is sufficiently close and the assumptions hold, the number of correct digits may grow rapidly across iterations.

“Quadratic convergence” refers to how the approximation error behaves, not necessarily that the original equation is a quadratic polynomial.

Exact error identity for positive square-root iteration

Let a=√2 and take any positive x. Define one Newton step N(x)=(x+2/x)/2.

Then N(x)−a=(x−a)²/(2x). This identity follows by substituting a²=2 and combining terms.

Because x>0 and the numerator is squared, the new estimate satisfies N(x)≥a. The formula explains why the first positive step lands at or above the root.

Worked example 6: derive the error identity

Start with N(x)−a=(x²+2−2ax)/(2x).

Since a²=2, the numerator becomes x²−2ax+a²=(x−a)².

Therefore N(x)−a=(x−a)²/(2x). When x is near a, the squared error in the numerator becomes very small.

A small residual is not identical to a small input error

The residual |f(x)| measures how nearly the equation is satisfied. The input error |x−r| measures how far the estimate is from the true root.

A small residual does not universally imply a small input error if the function is very flat or scaled by a tiny constant.

Therefore checking only |f(xₙ)| without understanding slope can give misleading confidence in a reported root.

Worked example 7: a scaled function with tiny residual

Let g(x)=10⁻⁹(x−1000). The exact root is x=1000.

At x=0, the residual is g(0)=−10⁻⁶, numerically small. Yet x=0 is 1000 units from the root.

The small coefficient made the function residual tiny across a wide range. Scale and slope matter when interpreting accuracy.

Why Newton steps can overshoot

A tangent is a local linear approximation. If the actual curve bends sharply, the tangent’s zero may be far from the curve’s true zero.

A large step can move outside the intended interval or toward a different root. That does not mean differentiation is incorrect; it means the local approximation was used too far.

Convergence depends on the function’s behaviour and the starting point, not only the update formula.

A zero derivative can stop the procedure immediately

For f(x)=x²−2, choose x₀=0.

The derivative f′(0)=0, so Newton’s formula would attempt dividing −2 by zero.

The update is undefined. Although the equation has valid roots ±√2, zero is an invalid starting point for this particular Newton iteration.

Worked example 8: repair an invalid start

For f(x)=x²−2, x₀=0 fails because f′(0)=0.

Choose instead x₀=1. The first update is (1+2)/2=1.5.

Continue to 17/12 and then 577/408, approaching the positive root. This is a valid example of changing a start after detecting an undefined derivative.

A negative start may approach a different root

The equation x²−2=0 has two real roots: +√2 and −√2.

If Newton’s square-root iteration starts with a negative nonzero x, the transformed iteration preserves a negative direction and approaches the negative root under the standard square-root setup.

This demonstrates that Newton’s method solves f(x)=0, not necessarily a particular root unless the starting strategy or domain specifies which one.

Worked example 9: begin at negative one and a half

Use x₀=−1.5 in xₙ₊₁=(xₙ+2/xₙ)/2.

Then x₁=(−1.5−1.3333…)/2=−1.416666…, followed by approximately −1.4142157.

The sequence approaches −√2. A problem asking for the positive square root should choose a positive starting guess or enforce the desired domain.

An iteration can cycle rather than converge

Consider f(x)=x³−2x+2, whose derivative is f′(x)=3x²−2.

At x₀=0, f(0)=2 and f′(0)=−2. Newton gives x₁=0−2/(−2)=1.

At x₁=1, f(1)=1 and f′(1)=1. Newton gives x₂=1−1=0. The method returns to zero, creating a two-point cycle.

Worked example 10: confirm the cycle

For f(x)=x³−2x+2, start x₀=0.

The next value is x₁=1. Then x₂=0, then x₃=1 by the same computations.

The sequence 0,1,0,1,… never converges to a root because f(0)=2 and f(1)=1 are both nonzero. This is a concrete failure case.

Starting values matter even for smooth polynomials

The cycling cubic is differentiable everywhere, and its Newton updates at zero and one are both defined.

Yet iteration fails to converge from that start. Smoothness alone does not guarantee global convergence.

A learner should not treat the algorithm as infallible merely because all the individual derivative calculations are valid.

A function can have several roots

Take f(x)=x³−x, which factors as x(x−1)(x+1).

Its roots are −1,0,1. Newton’s method can approach different roots depending on the starting estimate, and some starts may encounter unstable behaviour.

The problem must specify whether any root is acceptable or a particular root is sought. A bracket or domain restriction can help target the intended solution.

Worked example 11: evaluate a simple cubic’s roots

For f(x)=x³−x, substitute −1: −1−(−1)=0. Substitute zero: zero. Substitute one: 1−1=0.

Thus all three values satisfy the equation exactly.

An iterative solver returning one of them is not necessarily wrong, but an answer claiming it is the only real root would be incomplete.

Newton’s method compared with bisection

Bisection starts with a continuous function whose endpoint values have opposite signs, then repeatedly halves a bracket containing a root.

Newton’s method uses a local derivative to jump from a guess to a tangent intercept. It can converge much faster near a suitable simple root, but does not automatically preserve a bracket.

The Bukit Timah bisection and square-root lesson teaches the guaranteed bracket logic under its conditions; this article focuses on tangent-based iteration.

Worked example 12: one bisection step

For f(x)=x²−2, begin with [1,2], since f(1)<0 and f(2)>0.

Midpoint is 1.5. Its value f(1.5)=0.25>0, so the root remains in [1,1.5].

Bisection has cut the bracket width from one to 0.5. It has not used the derivative.

Worked example 13: compare one Newton step

Starting Newton at x₀=1.5 gives x₁≈1.4166667, much closer to √2 than the initial midpoint.

The exact positive root is about 1.41421356. Newton’s first step changes the guess using the tangent gradient.

Bisection, meanwhile, retains a certified sign-changing interval. Speed and robustness are different qualities and should both be considered.

A hybrid solver can combine both strengths

A safeguarded Newton method may maintain a bracket and attempt Newton steps while they are useful.

If a proposed tangent step falls outside the safe bracket or violates other checks, it can use a bisection step instead.

This strategy can preserve much of bisection’s reliability while benefiting from Newton’s local speed. The detailed algorithm depends on its exact safeguards and assumptions.

Worked example 14: reject an unsafe proposal

Suppose a continuous function has a verified root bracket [1,2], but an untrusted Newton update proposes x=5.

The value five lies outside the bracket. A safeguarded method may reject it and use midpoint 1.5 instead.

The midpoint keeps the next calculation inside the known interval; accepting five would discard the bracket guarantee.

Geometric meaning of the derivative

The derivative f′(xₙ) is the tangent slope at the current graph point.

If the slope is steep, the horizontal change needed to reach y=0 along that tangent may be moderate. If the slope is nearly flat while f(xₙ) is nonzero, the quotient f/f′ may be large.

This geometric picture explains why derivative magnitude affects stability in Newton’s update.

Worked example 15: a steep slope versus a flat slope

Suppose at one fictional point f(x)=1 and f′(x)=10. Newton’s correction is −1/10=−0.1.

At another point with f(x)=1 and f′(x)=0.01, correction is −100.

The same function residual can produce very different steps because the tangent slopes differ. Neither pair alone guarantees what happens after the step.

The iteration index labels guesses

Write x₀ for the starting guess, x₁ for the first updated guess and x₂ for the next.

The subscript is an iteration count, not an exponent. x₂ does not mean x squared, and xₙ₊₁ is not an algebraic power.

Clear notation makes it easier to audit successive approximations and diagnose a wrong formula entry.

Worked example 16: organise an iteration table

For x²−2 from x₀=1.5, record:

n=0: x₀=1.5000000, f(x₀)=0.25. n=1: x₁≈1.4166667, f(x₁)≈0.0069444. n=2: x₂≈1.4142157, f(x₂)≈0.0000060.

The table shows the guess and residual evolving together, rather than confusing successive guesses with successive polynomial powers.

Stop conditions need a precise purpose

A numerical computation cannot usually continue forever. A stopping rule decides when to report an approximate result.

Possible checks include |xₙ₊₁−xₙ| below a threshold, |f(xₙ)| below a residual threshold, or a certified error bound when available.

No single heuristic threshold proves correctness for every function. The appropriate tolerance depends on units, scaling and mathematical guarantees.

Worked example 17: a step-size stopping check

Suppose an invented solver uses threshold 0.001 for successive changes.

In the square-root iteration, |x₁−x₀|≈0.0833333 and |x₂−x₁|≈0.0024510, both larger than 0.001.

The next difference |x₃−x₂|≈0.0000021 is smaller. The step condition would be met after that update, but the root should still be checked by substitution if accuracy matters.

Small updates can occur far from a root

Consider a function whose derivative is very large relative to its residual, or an iterative method stagnating because of finite precision.

Successive guesses can become nearly equal even though the original equation is not adequately satisfied.

Therefore stopping based only on the last update can be misleading. A separate residual or bracket check improves the diagnosis.

Relative and absolute tolerances are different

An absolute tolerance such as 10⁻⁶ measures a difference in the original x-units.

A relative tolerance compares the difference to the scale of the iterate or root. It can be useful when solving equations whose roots have very different magnitudes.

Near zero, purely relative error requires care. Many practical algorithms combine absolute and relative criteria.

Worked example 18: two roots with different scales

An absolute error of 0.001 is tiny relative to a root around 1000 (about one part per million).

The same absolute error around a root 0.001 is a 100% relative difference.

The intended measurement or problem decides which accuracy description matters. “Accurate to 0.001” is not universally equivalent to “very small percentage error.”

Rounding should come after the iteration check

Rounding each iterate to two decimal places changes the algorithm from the intended high-precision calculation.

The near-root values 1.416666…, 1.414215…, 1.414213… may all display as 1.41 to two decimals, concealing meaningful improvement.

Keep enough intermediate precision. Report a rounded final value only after the stopping and verification requirements are met.

Worked example 19: premature rounding

If x₁≈1.4166667 is rounded to 1.42 before the next update, calculate (1.42+2/1.42)/2≈1.4142254.

The proper unrounded x₂ is about 1.4142157. The two differ because rounding altered the input to the next iteration.

Both are close to √2, but unnecessary rounding introduces avoidable approximation error.

Solving a cubic provides a second application

Take f(x)=x³−2x−5. We want a root near x=2.

At x=2, f(2)=8−4−5=−1. Derivative f′(x)=3x²−2 gives f′(2)=10.

Newton’s first update is x₁=2−(−1)/10=2.1.

Worked example 20: a second cubic iteration

At x=2.1, f(x)=2.1³−2(2.1)−5=9.261−4.2−5=0.061.

Derivative is f′(2.1)=3(2.1²)−2=13.23−2=11.23.

Newton gives x₂=2.1−0.061/11.23≈2.094568. The true nearby root is approximately 2.094551, so the second estimate is already close.

A small cubic residual is checkable

For the cubic x³−2x−5, substitute the approximate x≈2.094568 back into the original equation.

The result is very close to zero. It is not exactly zero because the displayed x-value is rounded.

A learner should record that the root is approximately 2.09455, with precision chosen appropriately rather than treating finite displayed digits as an exact algebraic identity.

Multiplicity can slow Newton convergence

A multiple root is repeated as a factor in the polynomial.

For example f(x)=(x−1)² has a repeated root at one. Its derivative f′(x)=2(x−1), which also vanishes at the root.

Newton’s ordinary iteration simplifies away from x=1 to xₙ₊₁=(xₙ+1)/2, giving linear convergence rather than the usual quadratic local behaviour for a simple root.

Worked example 21: repeated root iterations

Start at x₀=3 for f(x)=(x−1)².

The iteration gives x₁=(3+1)/2=2, x₂=(2+1)/2=1.5, x₃=1.25, x₄=1.125.

The error halves at each step. It converges, but noticeably more slowly than the squared-error behaviour near a simple root.

Newton’s method can be adapted for known multiplicity

If the root multiplicity m is known under appropriate conditions, a modified formula uses

xₙ₊₁=xₙ−m f(xₙ)/f′(xₙ).

For f(x)=(x−1)² with m=2 and a starting point x≠1, the modified update reaches x=1 in one step.

This is advanced numerical analysis. The learner should understand the ordinary method first and should not guess multiplicity without justification.

Worked example 22: multiplicity correction

For f(x)=(x−1)², f′(x)=2(x−1).

Use modified Newton with m=2: xₙ₊₁=xₙ−2(xₙ−1)²/[2(xₙ−1)].

For xₙ≠1, simplify to xₙ−(xₙ−1)=1. The update reaches the exact root in this special quadratic construction.

A graph can explain basin dependence

Different starting points may flow toward different zeros under repeated tangent steps.

The set of starts approaching a particular root is called a basin of attraction in iterative dynamics.

Some functions have complicated basins, cycles or divergent trajectories. A simple smooth graph does not guarantee every point will lead to the same solution.

Worked example 23: choose a start near the intended root

For f(x)=x²−2, the equation has roots ±√2.

Start x₀=1.5 to target the positive root; start x₀=−1.5 to target the negative root.

The initial sign helps select the intended root for this special iteration. It is not a general theorem that signs always separate basins for every nonlinear equation.

Newton updates can move outside a restricted domain

Suppose a mathematical model is defined only for x>0, perhaps because an expression contains ln x.

An update proposing x≤0 would fall outside that domain, and evaluating the next function or derivative might be invalid.

A robust computation checks feasibility after each proposed step and uses a suitable alternative if the domain would be violated.

Worked example 24: a logarithmic domain boundary

Consider f(x)=ln x−1 for x>0. Its root is x=e, and derivative f′(x)=1/x.

Newton update is xₙ₊₁=xₙ−xₙ(ln xₙ−1)=xₙ(2−ln xₙ).

At a very large positive starting value with ln xₙ>2, the next iterate is negative. The method would leave the permitted domain, illustrating the need for safeguards.

A derivative may be expensive or unavailable

In many practical numerical problems, f can be evaluated from data or a simulation, while its derivative is difficult to obtain.

Finite-difference estimates can approximate a derivative, but they introduce numerical error and potential sensitivity to noise.

That leads to quasi-Newton, secant and other methods. The present lesson uses known derivatives so the core tangent logic remains visible.

Worked example 25: a finite-difference slope

For f(x)=x²−2, approximate f′(1.5) by [f(1.6)−f(1.4)]/(1.6−1.4).

Values are f(1.6)=0.56 and f(1.4)=−0.04, so estimated slope is 0.60/0.20=3.

The exact derivative f′(x)=2x gives f′(1.5)=3. The symmetric finite difference happens to be exact for this quadratic at the midpoint.

A secant method uses two previous guesses

The secant method replaces the analytical derivative in Newton’s update by a slope from two distinct recent function values.

A common formula is xₙ₊₁=xₙ−f(xₙ)(xₙ−xₙ₋₁)/[f(xₙ)−f(xₙ₋₁)], when the denominator is nonzero.

This approach does not need f′ explicitly. It still lacks an unconditional convergence guarantee and can fail when the divided difference is zero or the starting guesses are unsuitable.

Worked example 26: one secant step

For f(x)=x²−2, start x₀=1 and x₁=2.

Then f(x₀)=−1 and f(x₁)=2. The secant update is x₂=2−2(2−1)/(2−(−1))=2−2/3=4/3≈1.3333.

This is inside the original bracket but not yet especially accurate. Later secant steps refine it under this favourable function setup.

Newton versus secant versus bisection

Newton uses a derivative at one current point. Secant uses two recent function values to approximate a slope. Bisection uses a sign-changing bracket and midpoint.

Each method trades information requirements, speed and reliability differently.

A parent does not need the child to memorise every numerical method. The foundational skill is to identify what the next estimate assumes and how the answer can be checked.

Precision claims need the correct type of error

An approximate root can be evaluated by input error, function residual, relative error or an interval bound.

These are related but not interchangeable. A suitable derivative lower bound near a simple root can sometimes convert a residual bound into an input error bound.

Without that information, reporting an extremely small residual as a guaranteed six-decimal-place root may be unjustified.

Worked example 27: use a derivative lower bound

Suppose f is differentiable on an interval containing estimate x and root r, and |f′(t)|≥m>0 throughout that interval.

By the mean value theorem, |f(x)−f(r)|≥m|x−r|. Since f(r)=0, |x−r|≤|f(x)|/m.

This is a conditional error bound. It requires the stated derivative lower bound over the interval, not just one derivative evaluation at the estimate.

Square-root iteration has a particularly clean check

For positive x, f(x)=x²−2. If x lies above √2, then

x−√2=(x²−2)/(x+√2).

The denominator is positive, so a small residual can yield a small input error when its lower bound is known.

This is more informative than copying a calculator value without understanding what the error means.

Worked example 28: bound error after one step

Take x₁=17/12, for which residual x₁²−2=1/144.

Because √2>1, denominator x₁+√2>17/12+1=29/12.

Thus error x₁−√2<(1/144)/(29/12)=1/348≈0.002874. The actual error is about 0.002453, consistent with the bound.

A root’s unit comes from the variable

If x measures time, the root has time units. If x is a dimensionless coordinate, so is the root.

The function’s derivative has output-units per input-unit, and f/f′ therefore has input-units. Newton’s subtraction x−f/f′ is dimensionally consistent.

This is a valuable cross-check for an applied model. Terms with incompatible units cannot legally be subtracted.

Worked example 29: check Newton correction units

Suppose f(x) measures a fictional residual volume in litres and x measures minutes.

Then f′(x) has units litres/minute. The quotient f/f′ has units (litres)/(litres/minute)=minutes.

Subtracting that correction from x, measured in minutes, is dimensionally meaningful. No actual tank or flow system is described.

A very large update may still be algebraically defined

Newton’s update can produce a large finite number even when the derivative is nonzero.

A defined result is not the same as a sensible next guess. An advanced numerical solver may cap step lengths or use a line search.

The algorithm should not be described as guaranteed to improve the root estimate at every step. Counterexamples show otherwise.

Worked example 30: examine a small derivative

For f(x)=x²−2, take x₀=0.01. Then f(x₀)=−1.9999 and f′(x₀)=0.02.

Newton’s correction is −f/f′=+99.995, so x₁=100.005, much farther from √2 than the starting guess in absolute input distance.

The step is valid algebraically but a poor immediate approximation. Subsequent iterations may recover slowly from such an extreme value.

Local convergence is not global convergence

A theorem saying Newton converges when the starting point is sufficiently close to a simple root under smoothness assumptions is a local guarantee.

It does not promise success for every possible initial guess in the entire real line.

A student should avoid replacing “may converge quickly near a suitable root” with “always finds the root,” which is mathematically false.

A numerical solver should include an iteration limit

If a method cycles, diverges or stops improving, an implementation should not repeat endlessly.

An iteration cap gives a finite exit condition, after which the solver can report that convergence was not established.

This is a practical computational safeguard and a useful classroom lesson about algorithm design. Failing to converge is information, not permission to invent a final root.

Worked example 31: recognise a cycle from a table

For f=x³−2x+2 with start x₀=0, the iterate list is 0,1,0,1,0,1,…

If a table shows the same two non-root values repeating, further identical updates will reproduce the cycle.

A responsible conclusion is “Newton’s method cycles from this start; no root has been obtained.” Reporting 0.5 as an average would not solve f(x)=0.

A derivative check belongs before an iteration

Write f and f′ explicitly, then verify the derivative formula before substituting numbers.

For f=x³−2x−5, derivative is 3x²−2. Writing 3x−2 would change Newton steps and could silently spoil convergence.

The iterative method is only as correct as its underlying derivative. The Bukit Timah differentiation lesson provides the necessary foundations.

Common error 1: forget the derivative denominator

A pupil writes xₙ₊₁=xₙ−f(xₙ) instead of xₙ−f(xₙ)/f′(xₙ).

That update is not Newton’s method. Its step has the wrong scale and possibly the wrong physical units.

Repair the geometric derivation from the tangent line before attempting another numeric table.

Common error 2: use the old input in the new function value

After finding x₁, the next update needs f(x₁) and f′(x₁).

Reusing f(x₀) while updating x₁ applies inconsistent information and does not follow the stated recurrence.

Label each row n, xₙ, f(xₙ), f′(xₙ), xₙ₊₁ to make the dependence clear.

Common error 3: treat a single improvement as a guarantee

From x₀=1.5, the first square-root step improves the estimate. That favourable example does not prove every Newton step on every function improves accuracy.

The cyclic cubic and near-zero derivative examples provide counterexamples.

A sound lesson teaches conditions and failures alongside success.

Common error 4: use a root outside the intended domain

For x²−2=0, negative Newton starts can approach −√2.

If the question asks for the positive square root, the negative value is a mathematically valid equation root but not the requested one.

The solver must preserve the domain or target condition in the final answer.

Common error 5: interpret small residual as perfect accuracy

A small |f(x)| can arise because f is scaled by a small constant or is flat near the chosen input.

Substitute back, check slopes and use appropriate error bounds where available.

Do not turn a function value of 10⁻⁶ into an unqualified statement that the root is accurate to six decimal places.

Common error 6: round every iterate too early

Rounding to two decimals may make successive guesses appear equal even when the underlying calculation is improving.

Keep higher precision in intermediate steps and round the final value according to the required tolerance.

The exact 17/12 and 577/408 representations in the square-root example show how precise iteration can be retained.

Primary foundations: estimation before calculus

Younger students can locate √2 between one and two by comparing squares.

They can refine an estimate by testing values such as 1.4 and 1.5, without using a derivative.

The early goal is understanding bounds and approximate numerical values. Formal Newton updates belong later in the learning progression.

Secondary preparation: functions and gradients

Students should understand function notation, evaluating f(x), plotting roots and differentiating simple polynomials.

They should also know that f(x)=0 is a different question from finding f′(x) or the maximum of f.

A weak derivative foundation is a reason to revise earlier algebra rather than memorise an iteration formula.

Advanced numerical Mathematics and syllabus boundaries

Newton’s method is appropriate enrichment for learners progressing toward calculus, numerical analysis or computational Mathematics.

The SEAB 2027 G3 Mathematics and Additional Mathematics directory and 2027 A-Level syllabus directory provide official course pathways.

The inclusion of Newton iterations, safeguard methods or convergence proofs here does not mean every technique is compulsory for a Secondary Mathematics examination.

A twenty-minute family diagnostic

Minutes 1–5: Evaluate f(x)=x²−2 at one and two. Explain why the positive root lies between them.

Minutes 6–10: Derive xₙ₊₁=xₙ−f(xₙ)/f′(xₙ) from a tangent line.

Minutes 11–15: Starting x₀=1.5, obtain x₁=17/12 and x₂=577/408. Check that their residuals decrease.

Minutes 16–20: Test x₀=0 for the square-root function and the two-cycle start x₀=0 for x³−2x+2. Explain why Newton’s method is not guaranteed to converge from every start.

A seven-day revision route

Day one revises roots and function values. Day two sketches tangent lines. Day three differentiates simple polynomials.

Day four derives and applies Newton’s update for square roots. Day five records residuals and stopping criteria. Day six compares bisection with tangent iteration.

Day seven diagnoses an invalid derivative or a two-cycle and proposes a safer starting or bracket strategy. Understanding failure modes is part of mastering an algorithm.

Practice questions with checked answers

A. For f=x²−2, find f′. Answer: 2x.

B. Simplify Newton’s update for f=x²−2. Answer: (xₙ+2/xₙ)/2 for xₙ≠0.

C. Starting x₀=3/2, find x₁. Answer: 17/12.

D. Find x₂ from x₁=17/12. Answer: 577/408.

E. Why is x₀=0 invalid for f=x²−2? Answer: f′(0)=0, so the Newton update divides by zero.

F. For f=x³−2x−5 at x₀=2, find x₁. Answer: 2.1.

G. For f=x³−2x+2 starting x₀=0, list x₁,x₂. Answer: 1,0; the iteration cycles.

Frequently asked questions

What does Newton’s method calculate?

It iteratively estimates a root of f(x)=0 using the tangent slope at the current guess.

Why is the derivative part of the formula?

The tangent line’s gradient determines where its horizontal-axis intercept lies; that intercept becomes the next estimate.

Does Newton’s method always converge?

No. It can encounter a zero derivative, take a poor step, cycle or converge toward a different root.

Why is the square-root example so fast?

For a suitable positive start near the simple root, its error is approximately squared and scaled at each step, giving rapid local improvement.

Is a small function residual proof of a small root error?

Not without appropriate information about the function, such as a derivative lower bound near the root.

Is bisection always slower?

Bisection often converges less rapidly near smooth simple roots, but its bracket structure provides a valuable reliability property when the assumptions hold.

What if the starting point has derivative zero?

The ordinary Newton update is undefined. Change the starting strategy or choose another root-finding method.

Why shouldn’t I round each iteration?

Premature rounding changes the next input and can conceal convergence or cause stagnation.

Are the curves and measurements real Bukit Timah data?

No. All functions and numerical readings are original fictional teaching examples.

How can parents recognise understanding?

The learner can derive the formula, compute and verify successive iterates, explain the derivative restriction and identify a legitimate failure case.

Official context and connected Bukit Timah Mathematics

Consult the SEAB 2027 G3 Mathematics syllabuses and 2027 A-Level course directories for level-specific expectations. The numerical root-finding problems here are instructional enrichment rather than actual examination items.

Read Mathematics in Bukit Timah: Numerical Bisection and Square-Root Estimates, Differentiation and Tangent Gradients, Quadratic Inequalities and Sign Charts and Trapezoidal Numerical Area Estimation for related concepts with distinct objectives.

The immutable Clementi Secondary 1 Mathematics tutoring reference anchors first-principles explanation, valid algebra and independently verified answers.

The lasting habit is to define the root and target domain, calculate the correct derivative, update using the current iterate, monitor residuals and stopping conditions, and acknowledge rather than conceal nonconvergence.

Read the three companion Mathematics in Bukit Timah articles from this batch: Vector dot products, projections and directional work · Parabola focus, directrix and geometric loci · 3D space diagonals, cross-sections and spatial distance. Together they explain vector direction, iterative roots, conic loci and spatial geometry, with original worked Mathematics and explicit model assumptions.

Discover more from eduKate Singapore

Subscribe now to keep reading and get access to the full archive.

Continue reading