Zoom in on a differentiable function and its graph looks like its tangent line. So near x = a the tangent line is the best straight-line approximation to y = f(x). We use it to estimate values such as ∛8.05 by hand, and, in the language of differentials, to estimate how much a small error in a measurement affects a computed quantity.
01 · The linear approximation
The tangent line at (a, f(a)) is y − f(a) = f′(a)(x − a). Solved for y it is the linear approximation (linearization) of f at a:
L(x) = f(a) + f′(a)(x − a), and f(x) ≈ L(x) for x near a.
Example: f(x) = ∛x at a = 8. f(8) = 2 and f′(x) = (1/3)x−2/3, so f′(8) = (1/3)(8)−2/3 = (1/3)(1/4) = 1/12. Then L(x) = 2 + (x − 8)/12, and ∛8.05 ≈ L(8.05) = 2 + 0.05/12 ≈ 2.0041667 (true value 2.0041580).
02 · Choose a nearby point you know
L(25) = 2 + 17/12 ≈ 3.4167 from a = 8 is a poor estimate of ∛25 ≈ 2.9240, because 25 is far from 8. Choose a point near 25 where the cube root is easy: a = 27, where ∛27 = 3 and f′(27) = 1/27. Then L(x) = 3 + (x − 27)/27 = 2 + x/27, and L(25) = 2 + 25/27 ≈ 2.9259: much better.
03 · Differentials
Move from x to x + Δx. The actual change in y is Δy = f(x + Δx) − f(x). The differential dx is an independent variable (we take dx = Δx), and the differential of y is
dy = f′(x) dx,
the change along the tangent line. For small dx, dy ≈ Δy. Example: y = x² + 7, so dy = 2x dx. At x = 1 with dx = Δx = 0.001: dy = 2(1)(0.001) = 0.002, while Δy = (1.001)² + 7 − 8 = 0.002001.
04 · Propagated error
If a quantity x is measured with a possible error dx, a quantity computed from it, q = f(x), has propagated error approximately dq = f′(x) dx. The relative error is dq/q (often given as a percentage).
Guided problems · Linear Approximations and Differentials
Original problems on linearization, differentials, and error. Open a hint if you need one, check your answers, then try another set.
Check your understanding · Linear Approximations and Differentials
Answer each question on paper first, then open the answer to check yourself.
Find the linear approximation of f(x) = ∛x at a = 8 and use it to estimate ∛8.05.
Show answer
f(8) = 2 and f′(8) = (1/3)8−2/3 = 1/12, so L(x) = 2 + (x − 8)/12. Then ∛8.05 ≈ 2 + 0.05/12 ≈ 2.004167.
Why is L(25) from a = 8 a poor estimate of ∛25, and what is better?
Show answer
25 is far from 8, where the tangent line has drifted away from the curve. Using a = 27: L(x) = 3 + (x − 27)/27, so ∛25 ≈ 3 − 2/27 ≈ 2.9259 (true value 2.9240).
For y = x² + 7, compare dy and Δy at x = 1 with dx = 0.001.
Show answer
dy = 2x dx = 0.002 and Δy = (1.001)² − 1 = 0.002001. They differ by (dx)² = 0.000001.
