The Newton-Raphson method finds a root of f(x) = 0 by repeating x₍ₙ₊₁₎ = xₙ − f(xₙ)/f′(xₙ) from a starting guess, and near a simple root it roughly doubles the number of correct digits each step. The bisection method is slower but cannot fail: it halves an interval [a, b] where f changes sign until the interval is as small as you need. Most numerical methods papers ask you to run one or both, show an iteration table, and say when to stop.
Below, both methods solve the same equation so you can compare them line by line, followed by root formulas, a transcendental equation, a case where Newton slows down, and one where it runs away.
The two methods in one table
| Feature | Bisection | Newton-Raphson |
|---|---|---|
| Needs | f(a) and f(b) of opposite sign | a starting value x₀ and f′(x) |
| Update | c = (a + b)/2, keep the half with the sign change | x₍ₙ₊₁₎ = xₙ − f(xₙ)/f′(xₙ) |
| Convergence | linear: error halves each step | quadratic near a simple root |
| Always converges? | yes, for continuous f | no: bad x₀ or f′ ≈ 0 can fail |
| Error bound | after n steps, error ≤ (b − a)/2ⁿ | no simple a priori bound |
| Typical use | safe first bracket, or when f′ is awkward | fast polishing once you are close |
Regula falsi (false position) sits between them: it keeps a bracket like bisection but uses a straight line through the end points, and in many B.Tech syllabi it is taught in the same unit. Our engineering mathematics tutoring page explains how numerical methods sessions with Ajay Vatsyayan follow your own syllabus and past papers.
Example 1: bisection on x³ + 2x − 7 = 0
Step 1: find a bracket. f(1) = 1 + 2 − 7 = −4 and f(2) = 8 + 4 − 7 = 5. The signs differ and f is continuous, so a root lies in (1, 2).
Step 2: decide how many steps. To be sure of three decimal places we need error < 0.0005. After n steps the error is at most 1/2ⁿ, so we need 2ⁿ > 2000, which means n ≥ 11 (2¹¹ = 2048).
Step 3: iterate. Each row tests the midpoint c and keeps the half where the sign changes.
| n | a | b | c = (a + b)/2 | f(c) |
|---|---|---|---|---|
| 1 | 1 | 2 | 1.5 | −0.625000 |
| 2 | 1.5 | 2 | 1.75 | 1.859375 |
| 3 | 1.5 | 1.75 | 1.625 | 0.541016 |
| 4 | 1.5 | 1.625 | 1.5625 | −0.060303 |
| 5 | 1.5625 | 1.625 | 1.59375 | 0.235687 |
| 6 | 1.5625 | 1.59375 | 1.578125 | 0.086536 |
| 7 | 1.5625 | 1.578125 | 1.570313 | 0.012829 |
| 8 | 1.5625 | 1.570313 | 1.566406 | −0.023808 |
| 9 | 1.566406 | 1.570313 | 1.568359 | −0.005508 |
| 10 | 1.568359 | 1.570313 | 1.569336 | 0.003656 |
| 11 | 1.568359 | 1.569336 | 1.568848 | −0.000927 |
Step 4: state the answer. After 11 steps the root lies in (1.568848, 1.569336), so x ≈ 1.569 to three decimal places.
Notice that f(c) does not shrink steadily. Bisection ignores the size of f and only uses its sign, which is why it is reliable and also why it is slow.
Example 2: Newton-Raphson on the same equation
Step 1: set up the formula. f(x) = x³ + 2x − 7, f′(x) = 3x² + 2, so
x₍ₙ₊₁₎ = xₙ − (xₙ³ + 2xₙ − 7)/(3xₙ² + 2)
Step 2: choose x₀. From the bracket, x₀ = 1.5 is reasonable.
Step 3: iterate to six decimal places.
| n | xₙ | f(xₙ) | f′(xₙ) | x₍ₙ₊₁₎ |
|---|---|---|---|---|
| 0 | 1.500000 | −0.625000 | 8.750000 | 1.571429 |
| 1 | 1.571429 | 0.023324 | 9.408163 | 1.568949 |
| 2 | 1.568949 | 0.000029 | 9.384807 | 1.568946 |
| 3 | 1.568946 | 0.000000 | 9.384778 | 1.568946 |
Step 4: stop. Two successive iterates agree to six decimal places, so x ≈ 1.568946.
Three Newton steps did more than eleven bisection steps. Look at the errors: about 0.0025 after step 1, then 0.000003 after step 2. That squaring of the error is what "quadratic convergence" means.
Example 3: an iterative formula for a cube root
Exam questions often ask you to derive a Newton formula for √N or ∛N, then use it.
To find ∛30, solve f(x) = x³ − 30 = 0. With f′(x) = 3x²,
x₍ₙ₊₁₎ = xₙ − (xₙ³ − 30)/(3xₙ²) = (2xₙ + 30/xₙ²)/3
Start at x₀ = 3, since 3³ = 27 is close to 30.
| n | xₙ | x₍ₙ₊₁₎ |
|---|---|---|
| 0 | 3.000000 | 3.111111 |
| 1 | 3.111111 | 3.107237 |
| 2 | 3.107237 | 3.107233 |
| 3 | 3.107233 | 3.107233 |
So ∛30 ≈ 3.107233. The same idea with f(x) = x² − N gives the familiar square root rule x₍ₙ₊₁₎ = ½(xₙ + N/xₙ).
Example 4: a transcendental equation
Solve x + ln x − 2 = 0 correct to six decimal places.
f(x) = x + ln x − 2 and f′(x) = 1 + 1/x. Since f(1) = −1 and f(2) = ln 2 ≈ 0.693, a root lies in (1, 2). Start at x₀ = 1.
| n | xₙ | f(xₙ) | x₍ₙ₊₁₎ |
|---|---|---|---|
| 0 | 1.000000 | −1.000000 | 1.500000 |
| 1 | 1.500000 | −0.094535 | 1.556721 |
| 2 | 1.556721 | −0.000697 | 1.557146 |
| 3 | 1.557146 | −0.00000004 | 1.557146 |
Root: x ≈ 1.557146. A common slip here is using log base 10 on the calculator when the question means natural log. In engineering papers "log x" in a calculus context usually means ln x, but read the question.
Example 5: when Newton slows down (a repeated root)
Take f(x) = x³ − 3x + 2 = (x − 1)²(x + 2). The root x = 1 is a double root, so f′(1) = 0 as well. From x₀ = 2, Newton gives
2 → 1.555556 → 1.297907 → 1.155390 → 1.079562 → 1.040288 → …
The error roughly halves each step: linear convergence, no better than bisection. The fix, if you know the root has multiplicity m, is the modified formula
x₍ₙ₊₁₎ = xₙ − m f(xₙ)/f′(xₙ)
With m = 2 the iterates become 2 → 1.111111 → 1.001949 → 1.000001, and quadratic convergence is back.
When Newton runs away
Newton can also diverge. For f(x) = x e^(−x), whose only root is 0, the formula simplifies to x₍ₙ₊₁₎ = xₙ²/(xₙ − 1). Starting at x₀ = 2 gives 4, then 5.333, then 6.564, moving further away each step. The tangent at any x > 1 meets the axis further to the right.
Warning signs in an exam: f′(xₙ) close to zero, iterates that grow, or iterates that bounce between two values. The safe response is to find a sign-change bracket first, take a few bisection steps, then switch to Newton.
Where students lose marks
- No bracket shown. For bisection, write f(a) and f(b) with their signs before the first midpoint.
- Keeping the wrong half. Compare the sign of f(c) with f(a), not with the previous f(c).
- Stopping too early. "Correct to four decimal places" means iterate until two successive values agree to four places, then round. One step agreeing in three places is not enough.
- Rounding every intermediate value to the final accuracy. Carry at least two extra digits in the working, round only the answer.
- Calculator in degrees. For equations with sin x or cos x, radians are required. A whole column can be wrong from row one.
- Wrong derivative. Check f′ once at x₀ before filling the table. One algebra slip makes every row wrong, even though the method marks may survive.
Which method should you use?
- The question names the method: use it, and show the formula you are iterating.
- You have no idea where the root is: tabulate f at integers to find a sign change, then bisect.
- f′ is easy and you have a close starting value: Newton.
- The question asks how many iterations are needed in advance: bisection, using (b − a)/2ⁿ < tolerance.
- Newton is crawling: suspect a repeated root and try the modified formula.
Practice set
- How many bisection steps are needed to locate a root in [2, 3] with error below 0.0001?
- Run three bisection steps for e^(−x) − x = 0 on [0, 1]. Which interval contains the root after step 3?
- Use Newton-Raphson with x₀ = 0 to find the root of x³ − 5x + 1 = 0 near 0, to six decimal places.
- Use x₍ₙ₊₁₎ = ½(xₙ + 41/xₙ) with x₀ = 6 for two steps.
- Show that Newton's method for f(x) = 1/x − N gives x₍ₙ₊₁₎ = xₙ(2 − N xₙ), then find 1/7 starting from x₀ = 0.1.
Answers. (1) 14 steps, since 2¹⁴ = 16384 > 10000. (2) Midpoints 0.5 (f > 0), 0.75 (f < 0), 0.625 (f < 0), so the root is in (0.5, 0.625). (3) 0.200000, 0.201639, 0.201640, so x ≈ 0.201640. (4) 6.416667, then 6.403139. (5) 0.13, 0.1417, 0.142848, 0.142857, converging to 1/7 ≈ 0.142857.
What comes after root finding
Newton's method extends to systems of equations, where f′ becomes the Jacobian matrix of partial derivatives; the guide to partial derivatives and the gradient covers the calculus behind it. The same "iterate and watch the error" thinking runs through numerical solutions of ODEs, which build on the exact methods in our first-order differential equations guide. Within the same engineering maths paper, the Fourier series worked examples cover the other unit students most often ask about.
How Maths Bodhi helps
We set out iteration tables the way examiners mark them, practise choosing the method, and review your own attempts. If numerical methods is part of a back paper, see our step-by-step plan for clearing an engineering maths back paper. First-year students coming from JEE can also look at the IIT first-year maths support page.
Home tuition is available in Gurgaon (Gurugram). Students elsewhere in India and abroad learn with us online. You can book a free demo class.
Written and checked by the Maths Bodhi team (November 2026).
Frequently asked questions
- What is the Newton-Raphson formula?
- x₍ₙ₊₁₎ = xₙ − f(xₙ)/f′(xₙ), repeated from a starting value x₀ until successive values agree to the required accuracy.
- What is the difference between the bisection method and Newton-Raphson?
- Bisection halves a sign-change interval and always converges but slowly; Newton uses the derivative and converges much faster near a simple root, but can fail.
- How many iterations does the bisection method need?
- The smallest n with (b − a)/2ⁿ below the required tolerance.
- Why does the Newton-Raphson method fail?
- Usually because f′(xₙ) is close to zero, the starting value is far from the root, or the root is repeated, which makes convergence slow.
- What is the order of convergence of Newton-Raphson?
- Order 2 (quadratic) at a simple root, and only order 1 at a repeated root.
- When should I stop iterating in an exam?
- When two successive iterates agree to the number of decimal places asked for, then give the rounded value.
