Secant Methods

The Central Idea

Newton's method uses the derivative

to estimate the local slope.

The secant method replaces that derivative with the slope through two nearby points:

This produces a derivative-free root-finding iteration.

What Is a Secant Line?

A secant line passes through two points on a curve.

For the points

and

the secant slope is

From Secant Line to Root Estimate

The secant method uses the \(x\)-intercept of the secant line as the next approximation.

Using point-slope form through

we write

At the \(x\)-intercept,

Deriving the Secant Formula

Set \(y=0\):

Therefore,

Substitute the secant slope:

An Equivalent Form

The same formula is often written as

Both forms produce the same next iterate when the denominator is nonzero.

Two Starting Values Are Required

Unlike Newton's method, which usually begins with one initial guess, the secant method requires two starting values:

These determine the first secant line.

Afterward, the newest two iterates are used to generate the next one.

The Iterative Pattern

Starting with

compute

from \(x_0\) and \(x_1\).

Then compute

from \(x_1\) and \(x_2\), and continue:

Example: Approximating \(\sqrt2\)

To approximate \(\sqrt2\), solve

Choose

Then

First Secant Step

Apply

Therefore,

Second Secant Step

Now use

Since

the next iterate is

This gives

Continuing the \(\sqrt2\) Iteration

Further secant steps give approximately

Therefore,

A Secant Table for \(\sqrt2\)

Geometric Meaning

At each step, the method:

1. takes two recent points on the graph;
2. draws the secant line through them;
3. finds the secant line's \(x\)-intercept;
4. uses that intercept as the next approximation.

The method replaces tangent information with a two-point slope estimate.

Secant Method versus Newton's Method

Newton's method uses

The secant method approximates that derivative by

In this sense, the secant method is a derivative-free relative of Newton's method.

Why Avoid the Derivative?

Derivatives may be:

• unavailable in closed form;
• expensive to compute;
• complicated to implement;
• known only through sampled data;
• unreliable when the underlying function is itself numerical.

In such situations, secant slopes can be operationally attractive.

Secant Method versus Bisection

Bisection keeps a bracket and halves an interval.

The secant method generally does not preserve a bracket.

Bisection emphasizes reliability.

The secant method often emphasizes faster progress without requiring a derivative.

The Secant Method Is Usually an Open Method

The standard secant method updates two approximations but does not require the root to remain between them.

Therefore, like Newton's method, it is usually classified as an open method.

The next point may fall outside the interval determined by the previous two points.

Bracketing Secant Variants

Some root-finding methods combine a secant-style line with a maintained sign-changing bracket.

A major example is the false-position method, also called regula falsi.

It uses a secant intercept but keeps endpoint signs arranged so that a root remains bracketed.

False Position versus Standard Secant

Both methods begin with two points and use a secant-line intercept.

The difference is what happens afterward.

The standard secant method keeps the two most recent iterates.

False position keeps a sign-changing bracket.

The False-Position Formula

Suppose

The secant line through

and

crosses the \(x\)-axis at

Updating False Position

After computing \(c\):

1. if \(f(c)=0\), stop;
2. if \(f(a)f(c)<0\), replace \(b\) with \(c\);
3. otherwise, replace \(a\) with \(c\).

This preserves the sign-changing bracket.

Why the Secant Method Can Be Faster than Bisection

Bisection always uses the midpoint.

The secant method uses function values to estimate where the root is likely to lie.

If the function is close to linear over the relevant region, the secant intercept can land much closer to the root than the midpoint would.

Convergence Rate

Near a suitable simple root, the standard secant method typically converges faster than linearly but slower than Newton's method.

Its order of convergence is approximately

This number is the golden ratio.

What the Convergence Order Means

Linear convergence has order \(1\).

Newton's standard local convergence near a simple root has order \(2\).

The secant method lies between them:

It often gains accuracy efficiently without paying the cost of derivative evaluation.

Example: Solving \(x^3-x-2=0\)

Let

Choose

Then

First Secant Step for the Cubic

Compute

Continuing the Cubic Example

Further iterations produce values approximately

Therefore,

The Denominator Must Not Vanish

The secant formula contains

in the denominator.

Therefore, the update fails if

Geometrically, the secant line would be horizontal and would not have a finite \(x\)-intercept.

A Nearly Zero Denominator Can Be Dangerous

If

is very small, the secant slope may be near zero.

Then the correction term can become extremely large.

The next iterate may jump far away from the current region.

Distinct Starting Inputs Are Needed

The secant slope also contains

If the two inputs are identical, there is no two-point secant line.

Therefore, begin with

Starting Values Matter

Poor starting values can lead to:

• slow convergence;
• large jumps;
• oscillation;
• convergence to an unintended root;
• a nearly horizontal secant;
• divergence.

Graphs, tables, or brackets can help choose useful starting values.

Multiple Roots Can Slow Convergence

Like Newton's method, the secant method behaves best near simple roots.

When

and

the root is multiple and convergence may become substantially slower.

Stopping by Successive Difference

A common stopping condition is

This measures how much the newest estimate differs from the previous one.

Stopping by Residual

We may also require

This checks how closely the newest approximation satisfies the original equation.

Using Both Conditions

A more robust stopping rule may require

and

This checks both iterate stability and residual size.

Maximum Iteration Safeguard

The secant method is not guaranteed to converge from every pair of starting values.

A practical implementation should therefore impose a maximum count such as

If the tolerance is not reached, the method can report nonconvergence.

Secant Method in Pseudocode

set x0 = first initial guess
set x1 = second initial guess

repeat
    f0 = f(x0)
    f1 = f(x1)

    if f1 - f0 = 0
        stop with failure

    x2 = x1 - f1 * (x1 - x0) / (f1 - f0)

    if stopping condition is satisfied
        return x2

    x0 = x1
    x1 = x2
until maximum iterations are reached

A More Defensive Secant Algorithm

set x0 = first initial guess
set x1 = second initial guess

repeat
    f0 = f(x0)
    f1 = f(x1)
    denominator = f1 - f0

    if abs(denominator) is too small
        stop or switch methods

    x2 = x1 - f1 * (x1 - x0) / denominator

    if abs(x2 - x1) < x_tolerance
       and abs(f(x2)) < residual_tolerance
        return x2

    x0 = x1
    x1 = x2
until maximum iterations are reached

False Position in Pseudocode

choose a and b with f(a)f(b) < 0

repeat
    c = b - f(b) * (b - a) / (f(b) - f(a))

    if f(c) = 0
        return c
    else if f(a)f(c) < 0
        b = c
    else
        a = c
until stopping condition is satisfied

Unlike the standard secant method, this procedure maintains a bracket.

False Position Can Stagnate

False position preserves a bracket, but one endpoint can remain nearly fixed for many iterations when the function is strongly unbalanced across the interval.

This can make the method converge more slowly than expected.

Modified false-position methods are designed to reduce this problem.

When the Standard Secant Method Is Attractive

The standard secant method is especially useful when:

• derivatives are unavailable or costly;
• two reasonable starting values are known;
• faster convergence than simple bracketing is desired;
• function evaluation is relatively inexpensive;
• a root is expected to be simple.

When a Bracketing Method May Be Better

Bisection or false position may be preferable when:

• a guaranteed root interval is important;
• the starting information already provides opposite endpoint signs;
• the function behaves unpredictably away from the root;
• robustness matters more than local speed.

Hybrid Secant Strategies

Practical solvers often combine secant-style interpolation with bracketing.

A hybrid algorithm may:

1. maintain a valid root bracket;
2. attempt a secant or interpolation step;
3. accept it if it is safe;
4. fall back to bisection otherwise.

This approach combines speed with stronger safeguards.

Secant Slopes as Numerical Derivatives

The expression

is a finite-difference approximation to a derivative.

As the two inputs move close together, the secant slope may approximate the tangent slope.

This connects root finding with broader numerical differentiation ideas.

Common Error: Using Only One Starting Value

The standard secant method needs two distinct starting inputs.

Without both

and

the first secant slope cannot be formed.

Common Error: Reversing the Function-Value Difference

The numerator and denominator must be used consistently.

One reliable form is

Reversing only part of a difference changes the sign and gives an incorrect update.

Common Error: Forgetting to Shift the Pair

After computing \(x_{n+1}\), the next secant step should normally use

and

The oldest value \(x_{n-1}\) is dropped.

Common Error: Assuming the Root Remains Between the Iterates

Standard secant iteration does not preserve a bracket.

The new approximation may lie outside the interval between the previous two values.

A bracketing guarantee requires a method designed to preserve one.

Common Error: Dividing by a Tiny Function Difference

If

is extremely small, the secant update can become unstable.

A numerical implementation should detect this condition before performing the division.

Common Error: Rounding Too Early

Secant steps depend on differences between nearby values.

Aggressive early rounding can corrupt these differences and distort the estimated slope.

Keep adequate internal precision until the final result is reported.

Common Error: Stopping Without Checking the Original Equation

A small change between successive iterates does not by itself prove that the approximation satisfies

Always check the residual

before reporting the result.

A Reliable Secant Workflow

1. Rewrite the problem as \(f(x)=0\).
2. Choose two distinct starting values \(x_0\) and \(x_1\).
3. Evaluate \(f(x_0)\) and \(f(x_1)\).
4. Check that \(f(x_1)-f(x_0)\) is not zero or dangerously small.
5. Compute the next secant intercept.
6. Shift the pair forward to the two newest iterates.
7. Monitor the change between iterates.
8. Monitor the residual.
9. Stop when the required tolerance is reached or a failure condition occurs.
10. Verify the final approximation in the original equation.

Example: Full Workflow for \(\sqrt5\)

Solve

Choose

Then

The first secant step is

Continue the \(\sqrt5\) Workflow

Now use

Since

the next step gives approximately

Continuing gives

Therefore,

Practice 1: Write the Secant Formula

State the standard secant update for solving

Practice 2: First Step for \(\sqrt3\)

Let

with

Compute \(x_2\).

Practice 3: Shift the Pair

After \(x_2\) has been computed from \(x_0\) and \(x_1\), which two values are normally used to compute \(x_3\)?

Practice 4: Denominator Failure

What problem occurs if

Practice 5: Compare with Newton

What information does Newton's method require that the secant method avoids?

Practice 6: Compare with Bisection

Does the standard secant method guarantee that the root remains between the two most recent iterates?

Practice 7: Successive-Difference Test

Suppose

and

Does

hold?

Practice 8: Residual

For

and

compute the residual.

Practice 9: False Position

What feature does false position preserve that the standard secant method generally does not?

Practice 10: Geometric Interpretation

What geometric point becomes the next iterate in the standard secant method?

Practice Answers

Practice 1

Practice 2

We have

Therefore,

Practice 3

The next step normally uses

The oldest value \(x_0\) is dropped.

Practice 4

The denominator

becomes zero, so the standard secant update is undefined.

Practice 5

Newton's method requires derivative information

The secant method estimates slope from two function values instead.

Practice 6

No. The standard secant method does not automatically preserve a root bracket.

Practice 7

Since

the condition is satisfied.

Practice 8

Practice 9

False position preserves a sign-changing interval that brackets at least one root under the usual continuity assumptions.

Practice 10

The next iterate is the \(x\)-coordinate where the secant line through the two current curve points crosses the \(x\)-axis.

What to Remember

• The secant method approximates a root using two recent function values.
• It replaces the derivative in Newton's method with a secant slope.
• The method requires two distinct starting values.
• The standard update is \(x_{n+1}=x_n-f(x_n)(x_n-x_{n-1})/[f(x_n)-f(x_{n-1})]\).
• The denominator must not be zero or dangerously small.
• The standard secant method generally does not preserve a root bracket.
• Near a suitable simple root, convergence is typically superlinear with order about \(1.618\).
• False position uses a secant intercept while preserving a sign-changing bracket.
• Stopping conditions should monitor iterate changes and residuals.
• Hybrid methods can combine secant speed with bracketing reliability.

Where We Go Next

Bisection, fixed-point iteration, Newton's method, and secant methods all solve individual equations by generating increasingly useful numerical approximations. The next step is to broaden this perspective to equations that may involve several variables or several equations at once.

The next lesson, Numerical Solutions of Equations and Systems, develops a unified view of numerical equation solving and introduces iterative methods for coupled systems.