Directional Derivatives and the Gradient

In this lesson you’ll learn how to compute the rate of change of a function in any direction you choose, not just along the x or y axes. The directional derivative generalizes partial derivatives, and the gradient vector turns out to be the key to all of it.

Partial derivatives tell you the rate of change along the x-axis (holding y constant) and along the y-axis (holding x constant). But what if you want to know the rate of change in some other direction, like diagonally?

That’s what the directional derivative does.

The directional derivative of f at a point P in the direction of a unit vector u is

$$D_{\mathbf{u}} f = \nabla f \cdot \mathbf{u}$$

That’s it. Take the gradient, dot it with the direction you care about. The result is a number telling you how fast f changes as you move in that direction.

The unit vector part is important. The direction vector u must have length 1. If someone gives you a direction like ⟨3, 4⟩, you need to normalize it first by dividing by its magnitude: u = ⟨3/5, 4/5⟩.

The gradient ∇f is more than just a collection of partial derivatives. It has deep geometric meaning:

• It points in the direction of steepest ascent (where f increases fastest)
• Its magnitude is the maximum rate of increase
• The directional derivative is maximized when u points in the same direction as ∇f
• The directional derivative is zero when u is perpendicular to ∇f (along a level curve)
• The directional derivative is most negative when u points opposite to ∇f (steepest descent)

Example 1: Computing a directional derivative

Let f(x, y) = x² + y² at the point (3, 4). Find the directional derivative in the direction of u = ⟨1, 0⟩ (pure x-direction).

The gradient is

$$\nabla f = \langle 2x, 2y \rangle \implies \nabla f(3, 4) = \langle 6, 8 \rangle$$

The directional derivative is

$$D_{\mathbf{u}} f = \nabla f \cdot \mathbf{u} = \langle 6, 8 \rangle \cdot \langle 1, 0 \rangle = 6$$

This makes sense: moving in the pure x-direction, the rate of change is just the partial derivative with respect to x, which is 2x = 6 at x = 3.

Example 2: Maximum rate of change

At the same point (3, 4), the maximum rate of increase is the magnitude of the gradient

$$|\nabla f| = \sqrt{36 + 64} = \sqrt{100} = 10$$

This maximum occurs in the direction of the gradient itself. The unit vector in that direction is

$$\hat{\nabla f} = \frac{\langle 6, 8 \rangle}{10} = \langle 0.6, 0.8 \rangle$$

So the function increases fastest at a rate of 10 when you move in the direction ⟨0.6, 0.8⟩ (which points away from the origin, as expected for this bowl-shaped function).

gradient ∇f (steepest, D = 10) direction u (D = 6) direction v (D = 8)

The orange arrow is the gradient ∇f = ⟨6, 8⟩, pointing in the direction of steepest ascent (D = 10). The green arrow is the direction u = ⟨1, 0⟩ from Example 1 (D = 6). The purple arrow is v = ⟨0, 1⟩ (D = 8). The gradient always gives the largest directional derivative.

Example 3: Zero directional derivative

What direction gives a directional derivative of zero? Any direction perpendicular to the gradient. At (3, 4), the gradient is ⟨6, 8⟩. A perpendicular vector is ⟨-8, 6⟩ (swap and negate one component). Normalizing: u = ⟨-8/10, 6/10⟩ = ⟨-0.8, 0.6⟩.

$$D_{\mathbf{u}} f = \langle 6, 8 \rangle \cdot \langle -0.8, 0.6 \rangle = -4.8 + 4.8 = 0$$

Zero. Moving perpendicular to the gradient means moving along the level curve, where f doesn’t change. This is why the gradient is always perpendicular to level curves.

Directional derivatives and the gradient show up constantly:

• Machine learning uses gradient descent, which follows the negative gradient to minimize a loss function. The gradient tells you the steepest direction, and you step opposite to it.
• Game engines use the gradient of a heightmap to determine slope direction for character movement, water flow, and AI pathfinding. The directional derivative tells you the slope in any specific direction.
• Physics uses gradients to describe force fields: the electric field is the negative gradient of electric potential.
• Weather modeling uses temperature gradients to predict wind direction (air flows from high to low pressure, perpendicular to isobars).