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 | Numerical Methods, Bisection and Approximating Square Roots

—

by

eduKate Asia
Original mathematical editorial cover featuring two linear-system solution directions, one narrow and sensitive and one well-conditioned, illustrated through clean geometric axes.

What is the side length of a square with area exactly 2 m²? Its length is √2 metres, but that is not a terminating decimal. A calculator shows approximately 1.41421356. How can a student justify an approximation rather than merely copy the display? One method is to find an interval containing the answer, repeatedly halve it and explain how narrow the remaining interval has become.

This Mathematics in Bukit Timah guide uses fictional classroom designs and number problems to develop estimation, root-finding, the bisection method, iteration, convergence, error bounds and verification. Any square mat, garden, measurement, threshold or design parameter below is an invented teaching example, not a surveyed Bukit Timah feature. It is also not a substitute for actual construction tolerances or engineering calculations.

The quick example starts with x²=2. Since 1²=1 and 2²=4, the positive solution lies between 1 and 2. Test the midpoint 1.5: its square is 2.25, which is too large, so the answer lies between 1 and 1.5. Repeating this sign-based comparison produces increasingly precise bounds around √2.

Article contents

Original mathematical editorial illustration about the difference between a small residual and an accurate numerical solution; all Bukit Timah examples are fictional.

Quick answer: approximation needs a target and a check

A numerical method finds an approximate solution when an exact expression may be inconvenient, unavailable or unnecessary for the decision. It does not turn approximation into an exact result.

Before calculating, identify the equation, the allowed domain, the starting interval and the desired accuracy. Then determine how each iteration changes the estimate or shrinks the interval.

A result such as 1.414 is meaningful only when the learner can explain what it approximates and how the method establishes a suitable error limit.

Why use a Bukit Timah classroom story?

Bukit Timah’s familiar gardens and schools provide an easy setting for imagining square floor mats, measuring plans or learning materials. But the square of area 2 m² is a perfectly defined mathematical object, not a claim about a real school’s equipment.

The point is to distinguish a precise mathematical relationship from the numerical decimal chosen to represent its solution.

The same iterative method can be used for an unknown length, an intersection of curves or a numerical threshold, provided the method’s conditions hold.

Begin with a number-line bound

The positive solution to x²=2 is greater than one because 1²=1<2. It is less than two because 2²=4>2.

Therefore 1<√2<2. This is already useful information even without calculating a decimal.

The interval’s width is 2−1=1. A numerical method can narrow that interval while keeping the true root inside, giving stronger information with each valid step.

Introduce a function whose root we need

Rewrite x²=2 as f(x)=x²−2=0. A root is an x-value where f(x)=0.

At x=1, f(1)=−1. At x=2, f(2)=2. The function is continuous on [1,2], and its values at the endpoints have opposite signs.

That sign change, together with continuity, guarantees at least one root in the interval by the intermediate value theorem. In this example, the nonnegative square function is strictly increasing, so the positive root is unique.

Worked example 1: the first midpoint

The midpoint of [1,2] is (1+2)/2=1.5. Evaluate f(1.5)=1.5²−2=2.25−2=0.25.

The value is positive, while f(1) is negative. Thus a sign change remains between 1 and 1.5, and the root lies in [1,1.5].

The interval width falls from one to 0.5. The original right endpoint two can now be discarded without losing the root.

Why do signs matter more than which number “looks close”?

The bisection method does not choose a half merely by asking whether the midpoint visually resembles the answer. It relies on a sign change for a continuous function.

If f(a) and f(m) have opposite signs, keep [a,m]. Otherwise, if f(m) and f(b) have opposite signs, keep [m,b]. If f(m)=0, the root has been found exactly.

This logic carries the guarantee forward. A student who always moves the lower endpoint upward may eventually discard the true root.

Worked example 2: the second midpoint

The current interval is [1,1.5]. Its midpoint is 1.25. Calculate 1.25²=1.5625, so f(1.25)=−0.4375.

The sign is negative. Since f(1.5)=+0.25, the root remains in [1.25,1.5].

The width halves again to 0.25. The method does not require knowing √2 in advance; the endpoint signs determine the next interval.

Worked example 3: narrow the range a third time

From [1.25,1.5], midpoint is 1.375. Its square is 1.890625, so f(1.375)=−0.109375.

The sign change therefore lies between 1.375 and 1.5. The retained interval is [1.375,1.5], with width 0.125.

Each successful bisection reduces the retained interval width by exactly one-half. That predictable reduction is one reason the method is attractive for teaching numerical error control.

A bisection table records reasoning

A useful table contains the lower endpoint, upper endpoint, midpoint, midpoint function value and chosen next interval. It should not record only a list of decimals.

The first three tested midpoints are 1.5, 1.25 and 1.375. Their function values are +0.25, −0.4375 and −0.109375.

A student who can reconstruct the next interval from these signs understands the algorithm better than one who memorises the emerging digits of √2.

Worked example 4: the fourth and fifth steps

The midpoint of [1.375,1.5] is 1.4375. Its square is 2.06640625, so f(1.4375)=+0.06640625. Keep [1.375,1.4375].

The next midpoint is 1.40625. Its square is 1.9775390625, so f(1.40625)=−0.0224609375. Keep [1.40625,1.4375].

The new interval width is 0.03125. Every step is justified by an actual sign calculation, not a guessed movement toward a calculator’s stored square-root value.

Worked example 5: two more steps

From [1.40625,1.4375], midpoint is 1.421875. Its square is 2.021728515625, so the function value is positive. Keep [1.40625,1.421875].

The midpoint of that interval is 1.4140625. Its square is 1.99957275390625, giving negative f value approximately −0.00042724609375. Keep [1.4140625,1.421875].

The retained interval now has width 0.0078125. The true √2 remains within it.

The interval width gives a defensible error bound

Suppose a root is known to lie inside an interval [a,b] of width w=b−a. If we use the midpoint m=(a+b)/2 as our estimate, its distance from any point in the interval is at most w/2.

Therefore the midpoint estimate has an absolute location error no greater than half the retained interval width, assuming the root containment is valid.

This is not a statistical confidence interval. It is a deterministic consequence of the root lying inside the stated bracket.

Worked example 6: error after seven halvings

Starting width is one. After seven successful halvings, the width is 1/2⁷=1/128=0.0078125.

A midpoint estimate from that final interval is therefore within 0.00390625 of the true root. The midpoint of [1.4140625,1.421875] is 1.41796875.

It is an approximation with a justified absolute error limit, though it is not yet enough to guarantee a particular two-decimal rounding because the bracket extends across the rounding boundary 1.415.

Small error bounds do not automatically guarantee decimal rounding

The actual value √2 is about 1.41421356 and rounds to 1.41 to two decimal places. But the bracket [1.4140625,1.421875] includes numbers that would round to 1.41 and others that would round to 1.42.

Even though the bracket is narrow, it crosses the exact halfway boundary 1.415. Thus the bracket alone does not prove the two-decimal result.

A student should check whether the entire interval lies inside the relevant rounding region instead of relying only on the number of calculations performed.

Worked example 7: certify rounding to two decimals

Continue the bisection process from [1.4140625,1.421875]. Test 1.41796875, then 1.416015625, then 1.4150390625; each is greater than √2 because its square exceeds two, so it becomes the upper endpoint.

Next test 1.41455078125, which is also above √2. The retained interval is now [1.4140625,1.41455078125].

Every number in that bracket rounds to 1.41 to two decimal places under the usual rule. The approximation is now supported by a rounding certificate rather than a guessed display value.

Bisection gives a predictable number of iterations

If an initial interval has width W, then after n midpoint halvings the retained width is W/2ⁿ. For W=1, ten halvings yield width 1/1024≈0.0009765625.

A midpoint estimate from that bracket has error at most half its width, approximately 0.00048828125.

This gives a stopping plan before starting. To make the midpoint error at most a chosen tolerance ε, find n large enough that W/2^(n+1)≤ε.

Worked example 8: choose an iteration count

Suppose the initial bracket width is one and the target midpoint absolute error is at most 0.001.

We need 1/2^(n+1)≤0.001, equivalently 2^(n+1)≥1000. Since 2¹⁰=1024, n+1=10 is enough, so n=9 midpoint halvings suffice under the width bound.

At n=9, the final width is 1/512≈0.001953125 and midpoint error at most 1/1024≈0.0009765625, below the requested threshold.

Continuous sign change is essential

The intermediate value theorem applies to a continuous function over the full closed interval. If a function jumps or is undefined inside, a sign change need not indicate a root.

For example, f(x)=1/x has f(−1)=−1 and f(1)=1, yet f is undefined at zero and never equals zero for any nonzero x.

The signs differ, but there is no root in [−1,1]. Applying bisection blindly would eventually encounter an undefined midpoint. The continuity condition prevents that mistake.

Worked example 9: a sign change without a root

Take f(x)=1/x on the interval from −1 to 1, excluding x=0 because it is undefined.

Its endpoint values have opposite signs. But solving 1/x=0 has no solution.

The failure is not a contradiction of the intermediate value theorem; its continuity assumption is violated. A good student checks that the function is defined and continuous throughout the candidate bracket.

A root may exist without a sign change

Consider f(x)=(x−2)². It equals zero at x=2 and is positive at other real x-values.

At x=1, f=1. At x=3, f=1. Both endpoint signs are positive despite a root between them.

Therefore the absence of opposite endpoint signs does not prove the absence of roots. Standard sign-change bisection cannot certify the root using this bracket, even though the function is continuous.

Worked example 10: an even-multiplicity root

For f(x)=(x−2)² and interval [1,3], f(1)=1 and f(3)=1. The usual opposite-sign starting condition is not satisfied.

But evaluating x=2 gives f(2)=0 exactly. A solver might find the root through factorisation, graph inspection or another procedure.

The distinction is important: a numerical method has conditions under which it guarantees success, not an obligation to succeed on every equation regardless of structure.

The function can have more than one root

Consider q(x)=x²−5x+4=(x−1)(x−4). Its roots are x=1 and x=4.

At x=0, q(0)=4; at x=5, q(5)=4. The endpoint signs agree even though two roots lie inside.

A bracket covering multiple roots can fail the simple sign-change test. When using bisection, choose an interval in which the method’s assumptions are established for the root of interest.

Worked example 11: isolate one of two roots

For q(x)=x²−5x+4, evaluate q(2)=4−10+4=−2 and q(5)=25−25+4=4.

There is a sign change on [2,5], and continuity guarantees a root there. In fact the relevant root is x=4.

The other root x=1 lies outside this bracket. Choosing a suitable interval can help isolate the particular solution requested.

What is iteration?

An iterative method starts from an initial guess or bracket and repeatedly applies a rule to obtain a refined result.

Bisection repeats midpoint testing. Other methods may update one numerical guess directly, without maintaining two endpoints.

An algorithm should specify its update, starting value and stopping condition. “Keep trying until it looks right” is not a mathematical specification.

A second approach: a square-root iteration

For a positive target number a, a classic iteration for √a is x_(n+1)=(x_n+a/x_n)/2, provided x_n>0. It is a special case of Newton’s method.

For a=2, start with x₀=1.5. The next estimate is half the sum of 1.5 and 2/1.5: x₁≈1.41666667.

This can converge rapidly near the positive root. However, its mathematical behaviour depends on the formula and starting conditions; not every arbitrary iteration converges.

Worked example 12: two Newton-style updates

Start x₀=1.5 and update x_(n+1)=(x_n+2/x_n)/2.

The first update is x₁=(1.5+1.333333…)/2=1.4166666667. The second is approximately x₂=1.4142156863.

Both are closer to √2 than the starting guess in this example. A third update gives approximately 1.4142135624. The rapid improvement contrasts with the steady halving of bisection.

Why Newton’s method is not the same guarantee as bisection

Bisection maintains an interval containing a root when continuity and opposite endpoint signs are valid. Its width reduction is predictable, though it can require many steps.

Newton’s method updates one estimate using local function behaviour or the special square-root formula. It can converge much faster near suitable roots, but may fail or leave a desired domain with poor starting conditions.

Speed and reliability are different criteria. A learner should know which properties are being claimed for the specific method rather than assume one universally dominates.

A general Newton update uses a derivative

For a differentiable function f(x), Newton’s method takes a current estimate x_n and updates it by x_(n+1)=x_n−f(x_n)/f'(x_n), provided the derivative is nonzero and the expression is defined.

Geometrically, the update finds where the tangent line at the current graph point crosses the horizontal axis. That intercept becomes the next estimate.

This is an upper-level extension. Pupils need not study calculus before appreciating the simpler bisection algorithm or the special square-root iteration.

Worked example 13: Newton on x-squared minus two

Let f(x)=x²−2. Its derivative is f'(x)=2x, so Newton’s rule gives x_(n+1)=x_n−(x_n²−2)/(2x_n).

For x_n≠0, simplify to (x_n+2/x_n)/2, the same square-root iteration used above.

Thus the shortcut has a mathematical derivation rather than being an unexplained recipe. The nonzero condition matters because dividing by 2x_n would be invalid at zero.

Convergence should be tested, not merely asserted

A sequence of estimates converges to L if its terms approach L as the iteration proceeds. Some update rules oscillate, diverge or settle at an unintended solution.

For example, iteration x_(n+1)=−x_n starting from x₀=1 produces 1,−1,1,−1,… . The sequence does not converge to one real limit.

A student should not believe that applying the same formula repeatedly necessarily improves accuracy. The update’s structure and starting value matter.

Worked example 14: an oscillating iteration

Take x₀=1 and rule x_(n+1)=−x_n. Then x₁=−1, x₂=1, x₃=−1 and so on.

The differences between successive terms remain two in absolute value. The sequence does not settle down.

This is a counterexample to the universal claim that repeated calculation automatically yields a better estimate. Iteration needs a convergence argument or appropriate checks.

Residual is not the same as root error

A residual is the function value f(x) at the current estimate. Root error is the distance between that estimate and an actual root.

A small residual can be useful evidence, but it does not always imply a small error in x. The scale and steepness of the function matter.

A solver who reports “the residual is nearly zero, so the estimate is exact” may be making an unsupported conclusion.

Worked example 15: a tiny residual far from the root

Let f(x)=0.0001(x−10), whose exact root is x=10. At guess x=0, f(0)=−0.001, a numerically small residual.

But the estimate is ten units away from the root. The small coefficient makes function values small even at distant x-values.

Therefore residual magnitude alone is insufficient as a universal accuracy guarantee. A valid bracket or additional information about the function’s slope can provide stronger control.

A large residual can still be meaningful

Suppose h(x)=1000(x−2), with root x=2. At estimate 2.001, the error in x is only 0.001, but residual h(2.001)=1.

Comparing residual values across different functions without accounting for their scale can mislead. The output unit and derivative influence the numerical residual.

The correct question is which quantity the problem needs to control: function output error, input location error or both.

Bisection bounds remain useful with calculator verification

A calculator may return a long decimal, but a proof-style check can test the nearby squares. If 1.4142²<2 and 1.4143²>2, then √2 lies between those numbers.

Those inequalities certify a four-decimal-place interval independently of trusting a screen display, provided the products were computed correctly.

This complements, rather than competes with, a calculator. Technology gives a candidate quickly; bounds explain what accuracy is justified.

Worked example 16: bracket with decimal squares

Calculate 1.4142²=1.99996164 and 1.4143²=2.00024449.

The first is below two, and the second is above two. Since x² increases for positive x, 1.4142<√2<1.4143.

The interval width is 0.0001. Its midpoint gives an absolute error at most 0.00005, under the mathematical monotonicity and bracket assumptions.

Estimation should begin before a detailed algorithm

For a square of area 50 m², side length is √50. Since 7²=49 and 8²=64, the side length lies between seven and eight metres.

A more targeted estimate can begin near 7.1, because 7.1²=50.41. The actual value is slightly below 7.1.

This simple reasonableness step helps catch factor-of-ten or wrong-operation errors even if the problem does not require a formal root-finding procedure.

Worked example 17: an invented fifty-square-metre mat

Suppose a fictional square mat must have mathematical area 50 m². Its side length satisfies x²=50, x>0.

We know 7<√50<8. Testing 7.1 yields 50.41, so the root is below 7.1. Testing 7.07 gives 49.9849, so the root is above 7.07.

Thus 7.07<√50<7.1. Those approximate numbers are part of a hypothetical geometry problem, not actual flooring specifications.

Binary search and bisection share a principle

Binary search for an item in a sorted list repeatedly cuts the remaining possibilities into roughly two halves. Numerical bisection similarly retains one half of a root-containing interval.

The two algorithms use different tests. Binary search compares an ordered item with the middle element; root bisection uses function signs and continuity.

The shared strategy is information gained by halving a search space. Students can understand that strategy well before learning formal numerical analysis.

Worked example 18: a sorted-number guessing game

A fictional teacher thinks of an integer from 1 to 100 and answers only whether each guess is too high or too low. A systematic strategy guesses near the middle of the remaining range.

Guess 50 first. If the answer is higher, discard numbers 1 through 50 and next search 51 through 100. Each reliable answer removes roughly half the candidates.

This is not identical to the numerical root proof, but it builds the same habit of keeping a clearly defined feasible interval after each test.

Significant figures and numerical accuracy are not identical

A student might report the bisection estimate 1.4142136 to eight decimal places even when the retained interval width is 0.01. The extra digits are not justified by that bracket.

Reporting a longer decimal string does not manufacture accuracy. The stopping criterion and verified interval determine the numerical claim that can be supported.

This connects to Mathematics in Bukit Timah | Rounding and Error Bounds, but root-finding focuses on controlling computational approximation, not only rounding of measured inputs.

Measurement error and numerical error can coexist

Suppose the target area of a real object is uncertain because its dimensions were measured approximately. Solving the ideal equation very accurately does not remove that input uncertainty.

A numerical algorithm can estimate the root of an assumed equation to many digits while the physical model still has larger errors from measurement or shape assumptions.

A complete conclusion may need both descriptions: computational tolerance under the equation and uncertainty in whether the equation matches reality.

A model’s domain must be checked

For x²=2, both +√2 and −√2 are real algebraic solutions. But if x represents the side length of a square, the physical domain is x>0.

Therefore the negative root is rejected for the geometry interpretation, not because the algebraic equation has only one real root.

The domain is part of the question. A numerical method should target the correct branch and check that each estimate remains relevant.

Worked example 19: a negative root is mathematically real

The equation x²−2=0 has two roots x=√2 and x=−√2. Both satisfy the algebraic equation.

For a square side length, only the positive one is valid. For a pure algebra problem asking for all real solutions, both must be reported.

This is the difference between solving an equation and interpreting a variable under its contextual restrictions.

A table of signs is more informative than a line of guesses

A student can list each midpoint, sign and retained bracket. That audit trail reveals exactly why a candidate was accepted or rejected.

If one midpoint sign was computed incorrectly, every subsequent bracket may be invalid. A teacher should inspect the first incorrect line, not merely the final approximation.

This aligns with small-group error diagnosis: correction is strongest when it repairs the concept that caused the wrong decision.

Primary progression: number lines, squares and bounds

Younger Primary pupils can use the squares 1,4,9,16 and so on to recognise intervals for square roots where appropriate to their readiness.

Place √2 between one and two on a number line, then estimate with a midpoint. The goal is conceptual understanding of approximate location, not immediately mastering continuous-function theorems.

Teachers can use square-area drawings to make the length–area relation visible before introducing a formal algorithm.

Lower Secondary: tables, inequalities and arithmetic checks

Students comfortable with decimals and directed numbers can calculate f(x)=x²−2, inspect its sign and narrow a bracket systematically.

They should label which endpoint changes and why. A graph of the function provides a second representation for the root as an x-intercept.

Exact syllabus placement varies. Numerical bisection may be enrichment even when square roots and coordinate graphs are already taught.

Upper Secondary: iteration and convergence

Learners with appropriate algebra and calculus prerequisites can derive Newton’s rule, study convergence properties and distinguish residual error from root error.

The extension should be explained as an additional mathematical method, not a universal requirement for every SEC Mathematics or Additional Mathematics paper.

The essential habit is comparing methods under stated conditions: stability, guaranteed containment, speed and the numerical precision actually required.

Four frequent mistakes tutors should diagnose

One student may choose the wrong half after a midpoint sign test. Another may assume a sign change proves a root even when the function is discontinuous. A third may confuse a tiny residual with tiny root error. A fourth may round the final value before the interval justifies it.

Those are different problems in logic, assumptions, interpretation and presentation. The corrective exercise should target the first invalid claim.

Students benefit from explaining the retained bracket aloud before writing the next calculation.

A twenty-minute home diagnostic

Minutes 1–5: Show that √2 lies between one and two by comparing squares. Define f(x)=x²−2.

Minutes 6–10: Test midpoint 1.5, keep [1,1.5]. Test 1.25, keep [1.25,1.5]. Ask why signs select those intervals.

Minutes 11–15: Calculate the next midpoint 1.375 and confirm the bracket [1.375,1.5]. Explain how width becomes 0.125 after three halvings.

Minutes 16–20: Compare f(x)=1/x, which changes sign across an undefined point, and f(x)=(x−2)², which has a root without an endpoint sign change. Ask what conditions the standard bisection guarantee actually needs.

A seven-day practice plan with retrieval

Day one focuses on square numbers and bounding √2. Day two evaluates function signs. Day three records a midpoint table. Day four calculates interval widths and error bounds.

Day five checks whether a decimal rounding is certified. Day six explores a second equation with a root in a new bracket. Day seven contrasts an example where bisection applies with one where its sign-change condition fails.

The student should be able to explain the method before using a calculator for a long decimal sequence.

Five challenge questions and verified answers

A. Where does the positive root of x²−3=0 lie between consecutive integers? Answer: between one and two, because 1<3<4.

B. For f(x)=x²−2 and [1,2], what is the first midpoint and its sign? Answer: 1.5, with f=+0.25.

C. What is the width after five halvings of an interval of width one? Answer: 1/32=0.03125.

D. Does f(x)=1/x have a root between −1 and 1 because its endpoint signs differ? Answer: no; it is undefined at zero, violating continuity on the interval.

E. What two real algebraic roots solve x²=2? Answer: ±√2. Only the positive root can describe a square’s side length.

Frequently asked questions

What is the bisection method?

It is a numerical root-finding algorithm that repeatedly halves an interval containing a root, choosing the half with a sign change for a continuous function.

Why must the function be continuous?

Continuity and opposite endpoint signs guarantee a root by the intermediate value theorem. A discontinuity can create a sign change without a root.

Is bisection guaranteed to find every root?

No. Its standard sign-change formulation needs suitable conditions and can miss even-multiplicity roots that do not change sign across a bracket.

Does a smaller residual always mean a better root estimate?

Not universally. A small function value can occur far from the root when the function has a very small scale or slope.

Why does interval width determine an error bound?

If the root is inside the retained interval, its distance from the interval’s midpoint cannot exceed half the interval width.

Is Newton’s method always faster and better?

It can converge rapidly under suitable conditions, but it does not offer the same simple bracketing guarantee for arbitrary starting values.

Are the square mats and gardens actual Bukit Timah measurements?

No. They are fictional models used to explain numerical approximation, not surveyed objects.

When should students stop iterating?

When a stated accuracy criterion is met and the solution remains valid under the model’s assumptions; sometimes a rounding boundary requires extra checks.

Which school levels should study numerical methods?

The foundational ideas of estimates and square roots suit earlier learning, while formal bisection, convergence and Newton’s method are advanced extensions based on course readiness.

Use the MOE Primary Mathematics syllabus and SEAB 2027 SEC G3 Mathematics syllabus listing to check level-specific content. The bisection and Newton extensions here are original enrichment explanations, not claims that every pupil is examined on these methods.

For precision and interval notation, revisit Mathematics in Bukit Timah | Measurement and Error Bounds. For the underlying quadratic curves, see Parabolic Arches and Turning Points. The immutable Clementi Secondary 1 Mathematics teaching reference anchors worked reasoning, targeted correction and independent verification.

The lasting habit is to state the equation and domain, verify the method’s conditions, keep a justified root-containing interval, stop to a meaningful tolerance and distinguish computational accuracy from the accuracy of the physical model.

Read the other three Mathematics in Bukit Timah articles in this batch: Network graphs, Euler trails and connections · Modular arithmetic, number codes and check digits · Functions, inverses and restricted domains. These four guides use different mathematical concepts, fictional teaching data and explicit limits on what each model proves.

Discover more from eduKate Singapore

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

Continue reading