Summary: Limits done carefully

Every derivative in this track has secretly been a limit: the value the rise-over-run ratio approaches as the interval shrinks. This lesson examines the limit itself, what “approaches” really means (made precise by the epsilon-delta idea), and how L’Hôpital’s rule rescues the awkward 0/0 forms the rate definition keeps producing. It bundles three short source chapters (the limit, the epsilon-delta definition, and L’Hôpital’s rule) into one. This is the scan-it-in-five-minutes version.

Core ideas

• Most limits are just “plug in.” For continuous functions the limit equals the value: lim (x->2) x² = 4. The interesting cases are the ones where plugging in fails.
• Indeterminate forms are the real content. 0/0 and ∞/∞ could approach any value or none, depending on how the top and bottom behave. The derivative itself is 0/0 if you set h = 0 directly, which is why limits deserve a lesson.
• What “approaches” means (epsilon-delta). Plain version: you can force f(x) as close to L as you want by making x close enough to a. Formal version: for any precision ε, there is a window δ around a that keeps f(x) within ε of L. It is a challenge (any ε) and a response (a δ that works), and it cleanly rules limits out too: sin(1/x) as x -> 0 oscillates forever, so no L works.
• L’Hôpital’s rule handles 0/0 and ∞/∞. Replace f/g with f'/g' (numerator and denominator differentiated separately, not the quotient rule), then retry, repeating if the form persists. Worked: lim (x->0) sin(x)/x = cos(0)/1 = 1 (proving the small-angle approximation); lim (x->0) (e^x-1)/x = 1; lim (x->0) (1-cos x)/x² = 1/2 (two passes); lim (x->∞) (ln x)/x = 0 (so x outpaces ln x).
• Why it works. Near a, each function is approximately its slope times the displacement; in the 0/0 case the constant terms vanish, the shared (x-a) cancels, and the ratio of derivatives remains. L’Hôpital keeps only the leading behavior, which is the seed of Taylor series.

What changes for you

The word “approaches” stops being vague and becomes a precise standard: a limit is L when you can force the output arbitrarily close to L by squeezing the input close enough to the point. That standard both confirms the easy limits and rules out the ones that do not exist, with one uniform test. And L’Hôpital gives you a fast, mechanical way through the 0/0 and ∞/∞ forms that the rate definition keeps throwing up: swap each function for its derivative and try again. In machine learning these ideas are rarely computed by hand but always underneath the guarantees: convergence analysis (does gradient descent actually settle?), continuous-time models (the step-size -> 0 limit of an ODE solver), and universal approximation theorems are all limit-based. With differentiation now fully built and on solid ground, the next lesson turns to the other half of calculus, integration, and the theorem that ties the two together.