MathLabs
TheoremProved

Directional derivative and steepest ascent along the gradient

Statement

If ff is differentiable at aa and ∇f(a)≠0\nabla f(a) \ne \mathbf{0}, then for any unit vector u\mathbf{u} (with ∥u∥=1\|\mathbf{u}\| = 1), the directional derivative satisfies Duf(a)=∇f(a)⋅u=∥∇f(a)∥cos⁡θD_{\mathbf{u}}f(a) = \nabla f(a) \cdot \mathbf{u} = \|\nabla f(a)\| \cos\theta, where θ\theta is the angle between ∇f(a)\nabla f(a) and u\mathbf{u}. Consequently, Duf(a)D_{\mathbf{u}}f(a) attains its maximum ∥∇f(a)∥\|\nabla f(a)\| when u=∇f(a)∥∇f(a)∥\mathbf{u} = \frac{\nabla f(a)}{\|\nabla f(a)\|}, and its minimum −∥∇f(a)∥-\|\nabla f(a)\| in the opposite direction.

Why is it true?

The dot product projects the gradient onto your chosen walking direction: you gain the full magnitude of the gradient only when you align perfectly with it, and zero change when you walk perpendicular to it along a contour line.

Proof sketch

Define the single-variable function g(t)=f(a+tu)g(t) = f(a + t\mathbf{u}). By definition of the directional derivative, Duf(a)=g′(0)D_{\mathbf{u}}f(a) = g'(0). Since ff is totally differentiable at aa, we have f(a+tu)−f(a)=∇f(a)⋅(tu)+o(∣t∣)f(a + t\mathbf{u}) - f(a) = \nabla f(a) \cdot (t\mathbf{u}) + o(|t|) as t→0t \to 0. Dividing by tt and taking the limit t→0t \to 0 yields Duf(a)=∇f(a)⋅uD_{\mathbf{u}}f(a) = \nabla f(a) \cdot \mathbf{u}.

By the geometric formula for the Euclidean inner product and the unit-length condition ∥u∥=1\|\mathbf{u}\| = 1, we obtain ∇f(a)⋅u=∥∇f(a)∥ ∥u∥cos⁡θ=∥∇f(a)∥cos⁡θ\nabla f(a) \cdot \mathbf{u} = \|\nabla f(a)\|\,\|\mathbf{u}\|\cos\theta = \|\nabla f(a)\|\cos\theta. Since −1≤cos⁡θ≤1-1 \le \cos\theta \le 1, the maximum value ∥∇f(a)∥\|\nabla f(a)\| is achieved uniquely at θ=0\theta = 0, where u=∇f(a)∥∇f(a)∥\mathbf{u} = \frac{\nabla f(a)}{\|\nabla f(a)\|}, and the minimum −∥∇f(a)∥-\|\nabla f(a)\| is achieved at θ=π\theta = \pi.

Topics that use this theorem

Step-by-step proofs

No step-by-step proof yet for this theorem.

References

  1. Jerrold E. Marsden, Anthony J. Tromba (2012). Vector Calculus
  2. James Stewart (2015). Calculus: Early Transcendentals
  3. Augustin-Louis Cauchy (1847). Méthode générale pour la résolution des systèmes d'équations simultanées