12.110 min read

Directional (U-) Derivative

Partial derivatives measure the rate of change of $f$ in the coordinate directions $\mathbf{e}_1, \ldots, \mathbf{e}_n$ . But what about an arbitrary direction? The directional derivative (or $U$ -derivative) measures the rate of change of $f$ in the direction of a unit vector $\mathbf{u}$ .

Formally, the directional derivative of $f$ at $\mathbf{a}$ in direction $\mathbf{u}$ is: $D_\mathbf{u} f(\mathbf{a}) = \lim_{t \to 0} \frac{f(\mathbf{a} + t\mathbf{u}) - f(\mathbf{a})}{t}$

This is the rate of change of $f$ along the line $\{\mathbf{a} + t\mathbf{u} : t \in \mathbb{R}\}$ passing through $\mathbf{a}$ in direction $\mathbf{u}$ . The partial derivatives are special cases: $D_{\mathbf{e}_i} f = \partial f/\partial x_i$ .

Formal View

Definition 12.1 — Directional Derivative

The directional derivative of

f: D \subseteq \mathbb{R}^n \to \mathbb{R}

\mathbf{a}

in the direction of a unit vector

\mathbf{u} \in \mathbb{R}^n

(

\|\mathbf{u}\| = 1

) is

D_\mathbf{u} f(\mathbf{a}) = \lim_{t \to 0} \frac{f(\mathbf{a} + t\mathbf{u}) - f(\mathbf{a})}{t}

provided the limit exists.

Some authors do not require $\|\mathbf{u}\|=1$ , but using unit vectors ensures the directional derivative measures rate of change per unit distance.

Why This Matters

Directional derivatives are how we reason about change in arbitrary directions, which is essential for optimization.

Finding the direction of steepest ascent/descent at a point
Computing rates of change in gradient-based optimization algorithms
Sensitivity analysis when inputs change in a prescribed direction

Learning Resources

Directional Derivatives

Khan Academy

Intuitive introduction to directional derivatives with geometric visualization.

11 min

Directional Derivatives and the Gradient

3Blue1Brown

Visual explanation of directional derivatives and their connection to the gradient.

16 min

Quiz

Question 1

The directional derivative $D_{\mathbf{e}_2} f(\mathbf{a})$ equals:

Question 2

The directional derivative $D_\mathbf{u} f(\mathbf{a})$ requires $\mathbf{u}$ to be a unit vector.

Common Mistakes

Forgetting to normalize the direction vector — directional derivatives with non-unit vectors scale by $\|\mathbf{u}\|$ .
Confusing the directional derivative (a scalar) with the gradient (a vector).
Thinking directional derivatives exist whenever partial derivatives exist — they require a stronger condition.