Local linearization is a method for approximating a complicated function with a simple linear one near a specific point. The core idea: if you zoom in close enough to any smooth curve, it starts to look like a straight line. Local linearization gives you the equation of that line so you can use it for quick, accurate estimates without dealing with the original function’s complexity.
The formula for a single-variable function f(x) near a point a is:
L(x) = f(a) + f'(a)(x − a)
This is the value of the function at your chosen point, plus the slope at that point times how far you’ve moved away from it. If that looks familiar, it’s the equation of the tangent line.
Why the Tangent Line Works as an Approximation
A tangent line touches a curve at exactly one point and runs parallel to the curve’s direction at that point. Right at the point of contact, the line and the curve share the same value and the same slope. Move a tiny distance away and they’re still nearly identical. Move far away and they diverge, sometimes dramatically.
This is what makes local linearization “local.” The approximation is only useful in a small neighborhood around the point you chose. How small depends on how much the original function curves, which is directly tied to its second derivative. A function that curves gently (small second derivative) stays close to its tangent line over a wider range. A function that curves sharply pulls away quickly.
The Error Bound
The gap between the true function value and the linear approximation follows a precise formula. For a point a distance h away from your chosen point:
|f(a + h) − L(a + h)| ≤ (M₂ / 2) · h²
Here M₂ is the largest value of the second derivative in the interval you’re looking at. Two things matter in this expression. First, the error grows with h², meaning it shrinks very fast as you get closer to your base point. Double your distance from a, and the error roughly quadruples. Second, M₂ captures how much the function curves. A nearly straight function (small M₂) gives a tiny error even at moderate distances, while a highly curved function demands you stay very close.
Connection to Taylor Series
Local linearization is the first-order Taylor polynomial. A Taylor series represents a function as an infinite sum of terms built from its derivatives at a single point. The zeroth-order term is just the constant f(a). Adding the first-order term, f'(a)(x − a), gives you the local linearization. Adding the second-order term, involving f”(a)(x − a)², gives you a quadratic approximation that hugs the curve more closely. Each additional term improves accuracy further from the base point, at the cost of more complexity. Local linearization stops after the first term because the whole point is simplicity.
Functions of Two or More Variables
The same idea extends naturally to multivariable functions. For a function f(x, y) near a point (x₀, y₀), the linearization becomes:
L(x, y) = f(x₀, y₀) + fₓ(x₀, y₀)(x − x₀) + f_y(x₀, y₀)(y − y₀)
Instead of a tangent line, you now have a tangent plane: a flat surface that just touches the curved surface of the function at one point and tilts in the same direction. The partial derivatives fₓ and f_y measure how steeply the surface slopes in the x and y directions, playing the same role that f'(a) plays in one dimension.
For three or more variables, you simply add another partial derivative term for each new variable. Using the gradient vector (the collection of all partial derivatives), the general formula becomes compact: L(x) = f(x₀) + ∇f(x₀) · (x − x₀). The dot product naturally accounts for the contribution of each variable’s slope.
An equivalent way to write the multivariable version focuses on changes rather than absolute values. If Δx and Δy are small changes from your base point, then the change in the function is approximately Δf ≈ fₓΔx + f_yΔy. This form is especially handy for estimating how a small shift in inputs affects the output.
The Pendulum: A Classic Physics Example
One of the most famous uses of local linearization is the small-angle approximation for a pendulum. The exact equation governing a swinging pendulum involves sin(θ), which makes it nonlinear and extremely difficult to solve analytically. But the linearization of sin(θ) near θ = 0 is simply θ itself. The Taylor expansion of sine is θ − θ³/6 + θ⁵/120 − …, and near zero, every term after the first is negligible.
Replacing sin(θ) with θ transforms the pendulum equation from an unsolvable nonlinear problem into a clean linear one that predicts simple harmonic motion. This is the version taught in introductory physics, and it works well for small swings. At 10 degrees (about 0.17 radians), sin(0.17) ≈ 0.1696, so the approximation introduces less than 0.3% error. At larger angles the cubic term starts to matter, and the simple formula breaks down.
How Engineers Use Linearization
Engineering control systems rely heavily on local linearization. Most real-world systems, from robotic arms to electrical generators, are governed by nonlinear equations. Designing a controller that keeps these systems stable is vastly easier when the equations are linear, because linear systems have well-understood, predictable behavior.
The standard approach is to pick an operating point (the state you want the system to stay near), linearize the equations around that point, and design a controller for the simplified version. This works as long as the system doesn’t wander too far from the operating point. For a two-wheeled balancing cart, for instance, linearization around the upright position lets engineers design controllers that keep it balanced during small disturbances. For larger deviations, more advanced techniques like feedback linearization can extend control to a wider operating range, but the starting point is almost always local linearization.
Marginal Analysis in Economics
Economists use local linearization constantly, often without calling it that. Marginal cost, marginal revenue, and marginal profit are all derivatives, and using them to estimate the effect of producing one more unit is a direct application of the linearization formula.
If C(x) is the total cost of producing x units, the exact cost of the next unit is C(x₀ + 1) − C(x₀). The marginal cost, C'(x₀), approximates this by treating the cost curve as locally linear. The approximation works because one unit is a small change relative to total production. At a factory making 10,000 widgets, the cost curve barely bends over one additional unit.
The same logic extends to percentage changes. If you know a function’s derivative and current value, the percentage change from a small input shift Δx is approximately 100 · f'(x)Δx / f(x). This is the foundation of elasticity calculations, where economists estimate how a small price change affects demand without re-evaluating the entire demand function.
Working Through an Example
Suppose you want to approximate √4.1 without a calculator. You know √4 = 2 exactly, so you linearize f(x) = √x at a = 4. The derivative is f'(x) = 1/(2√x), so f'(4) = 1/4. Plugging into the formula:
L(4.1) = 2 + (1/4)(4.1 − 4) = 2 + 0.025 = 2.025
The actual value of √4.1 is 2.02485…, so the linearization is off by about 0.00015, or less than 0.01%. The approximation is excellent because 4.1 is close to 4 and the square root function curves gently in this region (its second derivative at x = 4 is small).
Try the same thing further from the base point, say √9, using the same linearization at a = 4. You’d get L(9) = 2 + (1/4)(5) = 3.25, while the true answer is 3. That’s an 8% error, illustrating exactly why “local” is the key word.