Sumários

Lecture 4 - A. Frassino

14 novembro 2026, 11:00 • Antonia Micol Frassino


Introduction: constrained problems, an example in two variables, nomenclature.
Equality constraints: Lagrange multipliers, the Lagrangian function.
Numerical methods for equality constraints: Newton’s method on the Lagrangian, quadratic penalty.
Inequality constraints and KKT: minimizers on the boundary, sign and complementarity, the KKT conditions, an example where KKT works and one where it fails.
Optimality conditions: constraint qualification (LICQ), first- and second-order conditions.
Summary: how to solve a problem with KKT.

Lecture 3 - A. Frassino

13 novembro 2026, 19:30 • Antonia Micol Frassino


Introduction: multi-variables notation, least squares, some linear algebra.
Optimality conditions: convexity, minimizers, first- and second-order conditions, an example.
Line search methods: descent directions, steepest descent, choice of the step, Wolfe conditions.
Convergence: Zoutendijk’s theorem, rates of convergence, convergence of steepest descent.
Newton’s method: local convergence, Newton adapted to optimization, pros and cons.
Conclusion: optimality conditions, line search, and a comparison of the two methods (steepest descent and Newton).

Lecture 2

7 novembro 2026, 09:00 • Antonia Micol Frassino


Using Taylor expansions, we derived the conditions that a minimizer must satisfy. At a minimizer the first derivative is zero, and the second derivative tells a minimum from a maximum. We looked at what happens when the second derivative is also zero. We then introduced convex functions, whose graph lies below every chord, and showed that for them every local minimizer is also global. We also saw how to find the minimum of a function on a closed interval: compare the stationary points with the endpoints. Finally, we used MATLAB to solve one-variable problems numerically. We used the built-in functions fminbnd and fzero, applied Newton’s method to several functions and starting points, and ran gradient descent from several starting points to see how the result depends on where the method starts.

Lecture 1

6 novembro 2026, 17:30 • Antonia Micol Frassino


We introduced optimization as choosing the best value of some variables, and we turned a real problem into a mathematical model: the variables, the objective function to minimize or maximize, and the constraints. We defined global and local minimizers, and we saw that finding the candidates amounts to finding the zeros of the derivative. We also discussed when a minimizer is guaranteed to exist (the Weierstrass theorem), and where it can be: at a stationary point, at a point where the function is not differentiable, or at an endpoint of the interval