1. Newton–Raphson Method

Idea

One of the most powerful root-finding methods. At each step, replace with its first-order Taylor expansion (the tangent line) at the current point and solve that instead:

Algorithm


This is a fixed-point iteration with .

Example (extra) – ,

, , .
Convergence check: ✓

01140.75
10.750.1718752.68750.686047
20.6860470.0089412.4119790.682340
30.6823402.3967620.6823278039

The number of correct digits roughly doubles each step (2 → 5 → 10).

Newton–Raphson tangents, x₀ = 10.40.50.60.70.80.911.1-0.500.51x0x1x2xf(x)f(x)root


2. Convergence of Newton–Raphson

Theorem – Sufficient condition

If , is a root, and and have constant sign on , then choosing with

makes Newton–Raphson converge.

Theorem – Quadratic convergence (simple root)

With , the method is quadratically convergent if

For a simple root (), this limit is (extra).

Check on the example (extra): , , and the formula gives ✓

Theorem – Multiple root (linear convergence)

If is times continuously differentiable and is a root of multiplicity (so ), Newton is still locally convergent, but only linearly:

Example (extra)

(double root, ): , so → ✓

⚠️ Newton is sensitive to . It can diverge if isn’t close enough to the root, and it fails outright if (horizontal tangent).


3. Secant Method

Idea

When is hard or expensive to compute, replace it with a backward difference through the last two iterates:

Algorithm


It needs two initial points (no sign condition, unlike bisection).

Example (extra) – , ,

1010.5
210.50.636364
30.50.6363640.690052
40.6363640.6900520.682020
50.6900520.6820200.682326
60.6820200.6823260.6823278044

First step: .

⚠️ Extra: the secant method’s order is (superlinear). It’s slower per step than Newton, but each step needs only one new -evaluation and no derivative.


4. Comparison of the Methods

|xₙ − r| vs iteration for x³+x−1=0151015201e-121e-111e-101e-91e-81e-71e-61e-51e-41e-31e-21e-11e0iteration nerror (log scale)BisectionFixed point g=1/(1+x²)Newton (x₀=1)Secant (0,1)

MethodStrengthsWeaknesses
BisectionGuaranteed to converge; number of iterations for a given precision known in advanceSlow; the error doesn’t necessarily decrease every iteration
Fixed pointSimple to implement and understand; convergence guaranteed under conditions ()Slow in some cases; may not converge, or converge to an undesired root with a bad /guess
Newton–RaphsonFast (quadratic) convergence; extends to several variablesNeeds ; sensitive to the initial guess, can diverge if not close enough
SecantSimple; no derivative needed; acceptable (superlinear) speedSlower than Newton; can diverge with poorly chosen initial points
MethodOrderPoints neededNeeds Bracketing
Bisection1, 2 (sign change)noyes
Fixed point1 (usually)1no (but needs )no
Newton21yesno
Secant≈ 1.6182nono