17.26 min read

The Gradient as Normal Vector

We saw that the gradient $\nabla f(\mathbf{x}_0)$ and the isovectors at $\mathbf{x}_0$ are orthogonal: $\nabla f \cdot \mathbf{u} = Df \, \mathbf{u} = 0$ for every isovector $\mathbf{u}$ .

This means $\nabla f(\mathbf{x}_0)$ is perpendicular to the entire tangent space $T_{\mathbf{x}_0} S_k$ . We call it a normal vector to the isosurface $S_k$ at $\mathbf{x}_0$ .

Geometrically: the gradient stands straight out of the surface at each point, like a flagpole perpendicular to flat ground. The tangent plane lies flat, the gradient sticks out.

This normal vector relationship is what drives all of constrained optimization — the Lagrange condition will say that at an optimal constrained point, the gradient of $f$ must be parallel to the gradient of the constraint $g$ .

Formal View

Theorem 17.1 — Gradient is Normal to Isosurface

Let

f \in C^1

and

\mathbf{x}_0 \in S_k

with

Df(\mathbf{x}_0) \neq \mathbf{0}

. Then

\nabla f(\mathbf{x}_0)

is orthogonal to every isovector, i.e.,

\nabla f(\mathbf{x}_0) \perp T_{\mathbf{x}_0} S_k

. We call

\nabla f(\mathbf{x}_0)

a normal vector to

S_k

\mathbf{x}_0

Remark 17.1 — Tangent Plane

The set of all points

\mathbf{x}_0 + \mathbf{u}

for

\mathbf{u} \in T_{\mathbf{x}_0} S_k

forms the tangent plane to

S_k

\mathbf{x}_0

. The normal vector

\nabla f(\mathbf{x}_0)

is perpendicular to this tangent plane.

Why This Matters

The normal vector property is the geometric heart of Lagrange multipliers. An optimal constrained point must have $\nabla f$ parallel to $\nabla g$ — both normals to the same tangent plane.

Computing normal vectors to surfaces in computer graphics and physics
Finding the closest point on a surface to a given point (normal direction)
Gradient descent on manifolds: project gradient onto the tangent space
Reflection and refraction in optics computed using surface normals

Learning Resources

Lagrange multipliers introduction

3Blue1Brown

Shows visually how the gradient is perpendicular to level curves, setting up Lagrange multipliers.

13 min

Gradient and contour maps

Khan Academy

Geometric connection between gradient direction and isosurface normal.

10 min

Quiz

Question 1

The gradient $\nabla f(\mathbf{x}_0)$ is parallel to the tangent space $T_{\mathbf{x}_0} S_k$ .

Question 2

For $f(x,y) = x^2 + y^2$ , what is the normal to the circle $S_1 = \{(x,y) : x^2 + y^2 = 1\}$ at the point $(1/\sqrt{2}, 1/\sqrt{2})$ ?

Question 3

A curve $\mathbf{x}(s)$ stays on the isosurface $S_k$ and passes through $\mathbf{x}_0$ at $s = s_0$ . The velocity $\mathbf{x}'(s_0)$ must be:

Question 4

The gradient $\nabla f(\mathbf{x}_0)$ is always nonzero on an isosurface.

Common Mistakes

Saying the gradient is parallel to the isosurface (it is perpendicular).
Confusing the tangent plane (at a point on the surface) with the surface itself.
Thinking the gradient points along the isosurface toward higher values — it points away from the surface.