Linear Algebra
10.1110 min read

The Derivative

The derivative of ff at x0x_0 is the limit of the average rate of change as the interval shrinks to zero: f(x0)=limh0f(x0+h)f(x0)h.f'(x_0) = \lim_{h \to 0} \frac{f(x_0+h) - f(x_0)}{h}. It is a single number — the slope of the tangent line to the graph of ff at the point (x0,f(x0))(x_0, f(x_0)).

Geometrically: the derivative is the slope of the "best straight-line approximation" to the graph near x0x_0. If f(x0)>0f'(x_0) > 0, the function is increasing at x0x_0. If f(x0)<0f'(x_0) < 0, it is decreasing. If f(x0)=0f'(x_0) = 0, the tangent is horizontal — a candidate for a local extremum.

The derivative measures sensitivity: how much does ff change for a tiny change in input? A large f(x0)|f'(x_0)| means small changes in xx cause large changes in ff (sensitive). A small f(x0)|f'(x_0)| means ff is relatively flat (insensitive).

Formal View

Definition 10.10 — The Derivative
The derivative of ff at x0x_0 is
f(x0)=limh0f(x0+h)f(x0)h,f'(x_0) = \lim_{h \to 0} \frac{f(x_0 + h) - f(x_0)}{h},
provided this limit exists. We say ff is differentiable at $x_0$ when the derivative exists.

The derivative is the slope of the tangent line to the graph of ff at x0x_0. It equals the best linear approximation rate.

Interactive Visualization

Secant → Tangent Line

Why This Matters

The derivative is the central object of calculus — it enables optimization, rates of change, and local approximation.

  • Gradient computation in neural networks: the derivative of the loss w.r.t. each weight guides gradient descent.
  • Physics: the second derivative of position w.r.t. time is acceleration (F=maF = ma).
  • Finance: derivatives (financial) are named after this: their price depends on the derivative of the underlying.

Quiz

Question 1

The derivative f(x0)f'(x_0) is:

Question 2

If f(x0)>0f'(x_0) > 0, the function ff is increasing at x0x_0.

Common Mistakes

  • Thinking the derivative equals f(x0)/x0f(x_0)/x_0 — it is a LIMIT of differences, not a ratio of the function to its input.
  • Confusing f(x0)f'(x_0) (a number, the slope at one point) with the derivative function f(x)f'(x) (a function of xx).