Playing by Rules: Lagrange Multipliers

1. Introduction: The Fence

Usually, we minimize f(x). But what if we must stay on a path g(x) = c?

  • Example: Minimize cost (f), but satisfy nutritional requirements (g).
  • Example: Maximize Entropy (f), but probabilities must sum to 1 (g).

2. The Intuition

Imagine you are hiking up a hill f(x,y). You want to reach the highest point, but you must stay on a specific trail g(x,y)=0.

You walk along the trail.

  • If the trail cuts across the contour lines of the hill, you can still go higher by moving along the trail.
  • You stop when the trail runs parallel to the contour lines.

At this point, the gradient of the hill ∇f and the gradient of the trail ∇g are parallel!


3. The Lagrange Multiplier (λ)

The condition is:

∇f = λ∇g

We define a new function called the Lagrangian:

L(x, y, λ) = f(x, y) - λ(g(x, y) - c)

Finding where ∇L = 0 gives us the optimal point satisfying the constraint.


4. Interactive Visualizer: Constrained Climber

  • Background: Hill height map (f(x,y)).
  • Blue Line: Constraint (g(x,y) = x2 + y2 = 4, a circle).
  • Yellow Dot: Move it along the constraint circle.
  • Arrows:
    • Red: Gradient of Hill (∇f).
    • Green: Gradient of Constraint (∇g).

Find the point where the Red and Green arrows align (are parallel). That is the maximum!

Click/Drag on the Blue Circle. Align the Red and Green arrows.

5. Summary

  • Constraint: Restricts the search space.
  • Tangency: The optimal point is where the “objective slope” matches the “constraint slope”.
  • Lagrange Multipliers: The math tool to find these points analytically.

Next: Automatic Differentiation →