Approximating Values of a Function Using Local Linearity and Linearization

Long before digital computers could evaluate transcendental functions at the press of a button, mathematicians and scientists relied on elegant approximation techniques to carry out practical calculations. The idea that a smooth curve can be well represented by its tangent line in a small neighborhood is one of the oldest and most powerful insights of calculus. This principle, known as local linearity, underpins not only hand calculations in physics and engineering but also the numerical algorithms running on modern hardware. Its formal articulation as linearization crystallized over centuries of mathematical development, from the geometric methods of antiquity through the rigorous analysis of the 19th century.

The central question that linearization answers is deceptively simple: if you know the exact value of a function at one point and you know the derivative there, how accurately can you estimate the function at a nearby point without any additional computation? This question arises naturally whenever exact evaluation is expensive or impossible—for instance, estimating √4.02 without a calculator or predicting the behavior of a physical system slightly perturbed from equilibrium.

The foundation of linearization rests on a single geometric observation: when you zoom in sufficiently on a differentiable curve, it becomes indistinguishable from a straight line. This property is precisely what it means for a function to be differentiable at a point—the function possesses a well-defined tangent line there, and the curve hugs that tangent line ever more tightly as you restrict your view to smaller neighborhoods.

The diagram above illustrates the essential geometry of linearization. At the point of tangency (4, 2), the tangent line and the curve share both the same value and the same slope. As you move a small distance dx away from x = 4, the tangent line predicts a change of dy = f ′(4) · dx, while the actual change Δy on the curve is slightly different. The vertical red segment at x = 6 quantifies this discrepancy—about 0.051 units for a displacement of Δx = 2, which is not particularly small. Had we evaluated at x = 4.1 instead, the error would shrink to roughly 0.000125, demonstrating how rapidly accuracy improves as x approaches a.

The linearization formula follows directly from the definition of the derivative. Recall that differentiability at x = a means the limit that defines f ′(a) exists. Rearranging the difference quotient gives f(x) = f(a) + f ′(a)(x − a) + ε(x)(x − a), where ε(x) → 0 as x → a. Dropping the error term ε(x)(x − a) produces the linearization.

It is worth emphasizing that linearization is the first-degree Taylor polynomial of f centered at a. The Taylor expansion writes f(x) = f(a) + f ′(a)(x − a) + f ″(a)(x − a)²/2! + ⋯, so truncating after the linear term yields exactly L(x). The error in this truncation is bounded by the Lagrange remainder: |f(x) − L(x)| ≤ (M/2)|x − a|², where M is an upper bound on |f ″(c)| for c between a and x. This quadratic decay of the error explains why linearization works so well for small displacements.

A common AP Calculus BC exam task is to determine whether a linearization produces an overestimate or underestimate of the true function value. The key lies in the concavity of the function. If f is concave up (f ″ > 0) on an interval containing a, the tangent line lies entirely below the curve on that interval, so the linearization underestimates f(x) for x ≠ a. Conversely, if f is concave down (f ″ < 0), the tangent line lies above the curve and the linearization overestimates f(x). This geometric reasoning connects naturally to the second-derivative information already in the Lagrange remainder.

Let us approximate sin(0.1) using the linearization of f(x) = sin x centered at a = 0, and determine whether our estimate is an over- or underestimate.

Linearization is not an isolated technique; it is the first rung on a ladder of polynomial approximations that leads to Taylor and Maclaurin series, a major topic later in AP Calculus BC. The n-th degree Taylor polynomial P_n(x) centered at a includes terms through (x − a)ⁿ. The linearization is simply P₁(x). Each additional term reduces the error from O(|x − a|ⁿ⁺¹) to O(|x − a|ⁿ⁺²), so for small |x − a| the improvement is dramatic.

Beyond Taylor polynomials, the concept of local linearity reappears in multivariable calculus as the total differential and in differential equations as Euler's method, which repeatedly applies linearization to march along a solution curve step by step. Mastering linearization now builds intuition for all of these more advanced techniques.