Multi-Objective Optimization
Optimize several conflicting objectives at once, seeking the set of trade-off solutions rather than a single best point.
When one number is not enough
Real design rarely has a single objective. An engine might trade efficiency against weight; a plan might trade speed against risk. Multi-objective optimization minimizes a vector of objectives f(x) = (f1(x), ..., fk(x)) simultaneously. When objectives conflict, no single point is best on all of them, so the notion of optimality must change.
Pareto dominance
A solution x dominates y if it is at least as good on every objective and strictly better on at least one. A solution is Pareto optimal if nothing dominates it: you cannot improve one objective without worsening another. The set of all Pareto-optimal solutions is the Pareto set, and its image in objective space is the Pareto front.
Two solution philosophies
- Scalarization: combine objectives into one (weighted sum, epsilon-constraint) and solve repeatedly with different parameters to trace the front.
- Population methods: evolve a diverse set of solutions that approximate the whole front in one run, as in NSGA-II.
Choosing among trade-offs
Multi-objective optimization produces a set, not a single answer; a decision maker then selects a preferred trade-off. Preferences can be applied before optimization (fix weights), after (pick from the computed front), or interactively (refine preferences as the front is explored). Presenting the front makes the trade-offs explicit.
Quality metrics
Because the output is a set, its quality is judged by how close it is to the true front (convergence) and how well it spreads across the front (diversity). The hypervolume indicator, the volume of objective space dominated by the solution set, captures both in a single number and is widely used to compare methods.
Multi-objective methods lay out the trade-off surface among competing engineering goals so decision makers see the full range of achievable designs before committing.