Course overview
Most problems in life are optimisation problems: what is the best design for a racing kayak? How do you get the best return on your investments? What is the best use of your time in swot vac? What is the shortest route across town for an emergency vehicle? What are the optimal release rates from a dam for environmental flows in a river? Mathematical formulations of such optimisation problems might contain one or many independent variables. There may or may not be constraints on those variables. However, there is always an objective: minimise or maximise some function of the variable(s) subject to the constraints. This course examines nonlinear mathematical formulations, and concentrates on convex optimisation problems. Many modern optimisation methods in areas such as design of communication networks and finance rely on the classical underpinnings covered in this course. Topics covered are: One-dimensional (line) searches, direct methods, polynomial approximation, methods for differentiable functions; Theory of convex and nonconvex functions relevant to optimisation; Multivariable unconstrained optimisation, quasi-Newton Method, steepest descent methods, conjugate gradient methods; Constrained optimisation, including Karush-Kuhn-Tucker conditions and the Gradient Projection Method; Heuristics for non-convex problems, genetic algorithms.
Course learning outcomes
- Understand the complexities of, and techniques for solving, nonlinear optimisation problems.
- Apply suitable algorithms to one- or multi-dimensional optimisation problems.
- Understand the theoretical framework underlying the techniques presented in class.
- Implement computer code for the algorithms as studied in class and critically analyse and interpret the results.
- Demonstrate skills in communicating mathematics orally and in writing.
- Demonstrate the ability to investigate and analyse material related to the course.