12.2110 min read

Practical Minimization Method

When the domain $D$ is compact and $f$ is $C^1$ , a systematic method for finding the global minimum combines Fermat's theorem with boundary analysis:

Step 1: Find all interior critical points by solving $\nabla f = \mathbf{0}$ in the interior of $D$ . Step 2: Find all critical points on the boundary $\partial D$ (using Lagrange multipliers or parameterization of the boundary). Step 3: Evaluate $f$ at all critical points found in Steps 1 and 2. Step 4: The global minimum is the smallest value among all these candidates.

This method works because the EVT guarantees a minimum exists, and by Fermat's theorem, every interior minimum is a critical point. The boundary requires separate analysis since $\nabla f \neq \mathbf{0}$ at boundary minima is possible.

Formal View

Algorithm 12.2 — Practical Minimization on Compact Domain

To find

\min_{\mathbf{x}\in D} f(\mathbf{x})

for continuous

f

on compact

D

: 1. Find interior critical points: solve

\nabla f(\mathbf{x}) = \mathbf{0}

for

\mathbf{x} \in \text{int}(D)

2. Find boundary extrema: analyze

f|_{\partial D}

3. Evaluate

f

at all candidates 4. Return the minimizer

Why This Matters

This systematic method is the workhorse for constrained optimization in engineering, economics, and data science.

Quadratic programming: minimize a quadratic objective over a box or polytope
Portfolio optimization with investment constraints (budget, bounds on weights)
Control theory: optimal control on bounded input/state spaces

Learning Resources

Finding Global Extrema on Closed Domains

Khan Academy

Worked examples applying the systematic minimization method.

10 min

Constrained Optimization Introduction

MIT OpenCourseWare

MIT lecture on optimization over constrained domains.

48 min

Quiz

Question 1

When minimizing $f$ over a compact domain, which points must be considered as candidates?

Question 2

On a compact domain, every critical point in the interior is the global minimum.

Common Mistakes

Checking only interior critical points and ignoring the boundary.
Forgetting to evaluate $f$ at all candidates — just finding the critical points is not enough.
Not applying Lagrange multipliers correctly for boundary analysis when the boundary is curved.