The linearization looks like a new formula to memorize. It isn't — it's a line you already know, used in a new way. Here's the whole idea, built from the ground up, one step at a time.
Now watch it live. Here's a curve and its tangent at a. Slide a to move the tangent point; slide zoom to close in. Watch two things: as you zoom in, the curve becomes the line (local linearity), and the estimate L(x) snaps toward the true f(x). The gap between them is the error — and its direction is set by which way the curve bends.
Every linear approximation is this one formula. Read it as a set of directions: start where you know the value, then walk along the tangent's slope for however far you've stepped.
Zoom far enough into a smooth point and the curve is indistinguishable from a straight line — its tangent. That's what makes the estimate work at all.
The tangent line written as a function: L(x) = f(a) + f′(a)(x−a). Use it in place of f near a. Exact at a, approximate nearby.
Concave up: the tangent lies below, so L under-estimates. Concave down: tangent above, L over-estimates. The sign of f″(a) decides.
The same idea in change-language: dy = f′(x)\,dx estimates the output change Δy for a small input change dx — the tangent's rise over that step.
Six problems — build a linearization, estimate with it, and call the over/under. Try each before revealing.
Six fast checks. Pick an answer to see whether it's right and why — then work all six.
The step is measured from the anchor. Forgetting the −a shifts the whole line and wrecks the estimate. Always L(x)=f(a)+f′(a)(x−a).
The tangent only hugs the curve near the anchor. Use the value of a closest to your target — √4.1 from a=4, not a=1 — or the error balloons.
Concave up means the tangent is below the curve, so L under-estimates. It feels backwards — draw the bend and the tangent to be sure.
L(x) equals f(x) only at x = a. Everywhere else it's an approximation; say "approximately" and know there's an error.
dy = f′(x)dx is the tangent-line change (the estimate); Δy = f(x+dx)−f(x) is the true change. dy approximates Δy for small dx.
If f is decreasing at a, f′(a) < 0 and the linearization slopes down. Compute f′(a) with its sign and keep it.