Higher-Order Derivatives — Second, Third, and Beyond

What Are Higher-Order Derivatives?

Just as the first derivative f'(x) is the derivative of f(x), the second derivative f''(x) is the derivative of f'(x). You simply differentiate again. Notation: f''(x), d²y/dx², ÿ (Newton's double-dot). The nth derivative can be written f⁽ⁿ⁾(x) or dⁿy/dxⁿ.

Physical Meaning

Position s(t) → Velocity s'(t) → Acceleration s''(t). If position is measured in metres and time in seconds, the second derivative has units m/s² — metres per second squared, the unit of acceleration. This is Newton's second law: F = ma = m·s''(t).

Concavity and the Second Derivative Test

• f''(x) > 0: the curve is concave up (like a bowl) — the first derivative is increasing.
• f''(x) < 0: the curve is concave down (like a hill) — the first derivative is decreasing.
• f''(x) = 0: possible inflection point — where concavity may change.

Second Derivative Test for Extrema

• If f'(c) = 0 (critical point) and f''(c) > 0 → local minimum at c.
• If f'(c) = 0 and f''(c) < 0 → local maximum at c.
• If f''(c) = 0 → test is inconclusive (use first derivative test instead).

Example

f(x) = x³ − 3x. f'(x) = 3x² − 3. f''(x) = 6x. Critical points: f'(x)=0 → x = ±1. f''(1) = 6 > 0 → local min at x=1. f''(−1) = −6 < 0 → local max at x=−1.

The Full Hierarchy

Each derivative answers a more refined question about how a function changes. Position tells you where something is. Velocity (first derivative) tells you how fast it's moving. Acceleration (second derivative) tells you how fast the velocity is changing. Jerk (third derivative) tells you how fast the acceleration is changing — relevant in vehicle safety, where sudden changes in acceleration are felt as jolting. Snap (fourth derivative), crackle (fifth), and pop (sixth) are used in spacecraft trajectory design and precision robotics.

Beyond physics, higher derivatives measure the shape of curves with increasing precision. The second derivative tells you whether a curve is concave up or down. The third derivative tells you how fast the concavity is changing. Taylor series — which approximate any smooth function as a polynomial — use all derivatives up to order n, giving a complete local portrait of the function.

Computing Higher Derivatives — Full Examples

The Second Derivative and Curve Sketching

Concavity is one of the most important geometric properties of a curve, and the second derivative is how we measure it. A curve is concave up (shaped like ∪) where f''(x) > 0 — the slope is increasing. It is concave down (shaped like ∩) where f''(x) < 0 — the slope is decreasing. Points where the concavity switches are inflection points, where f''(x) = 0 and f'' changes sign.

Knowing concavity makes curve sketching precise. Without the second derivative, you know where a function rises and falls. With it, you know the exact shape of each arc — whether it bends upward or downward throughout each interval.

Second Derivative Test — Full Procedure

To classify critical points of f(x):

1. Find f'(x) and solve f'(x) = 0 for critical points c.
2. Compute f''(c) for each critical point.
3. f''(c) > 0 → local minimum (curve is concave up, so the critical point is at the bottom of a bowl).
4. f''(c) < 0 → local maximum (curve is concave down, critical point is at the top of a hill).
5. f''(c) = 0 → inconclusive. Use the first derivative test instead.

Why the Second Derivative Test Can Fail

The test fails when f''(c) = 0. Consider f(x) = x⁴: f'(0) = 0 and f''(0) = 0, yet x = 0 is clearly a local (and global) minimum. And f(x) = x³: f'(0) = 0 and f''(0) = 0, yet x = 0 is an inflection point, not an extremum. When f''(c) = 0, you must examine the sign of f' on either side of c, or examine higher-order derivatives.

Applications Beyond Calculus

In economics, the second derivative of a production function with respect to a factor input measures diminishing returns — the rate at which adding more labour or capital yields smaller and smaller gains. In machine learning, the Hessian matrix (the matrix of all second partial derivatives of the loss function) is used by second-order optimisers (Newton's method, L-BFGS) to achieve faster convergence than first-order gradient descent.