Higher-Order Derivatives

Single-variable calculus from first principles

If the first derivative f′ tells you the slope, what does the derivative of the slope tell you? That's the second derivative f″, and it measures how the slope is changing, which is the curve's concavity.

Just differentiate twice. For f(x) = x³: first f′ = 3x², then f″ = 6x. You can keep going (third, fourth derivatives) each one differentiating the last.

The sign of f″ tells you which way the curve bends. If f″ > 0 the curve is concave up: it cups upward like a bowl (∪), and the slope is increasing. If f″ < 0 it's concave down: it caps over like a dome (∩), and the slope is decreasing. Where the concavity flips is an inflection point.

Where this lives in MLThe second derivative is the 1-D seed of the Hessian matrix, the table of all second derivatives used in second-order optimisation (Newton's method) and in checking whether you've found a true minimum. Concavity is exactly convexity (next lessons): f″ ≥ 0 everywhere means a single global minimum and an easy optimisation landscape. And the second-order term is the curvature piece of a Taylor…
▶ Higher-Order Derivatives
← Implicit DifferentiationCritical Points →