Definition
An objective function is the scalar quantity an optimization procedure minimizes or maximizes over an allowed decision set. It encodes what the model treats as better, such as mean loss, tracking error or a risk-adjusted criterion. The objective is not meaningful without its units, data, constraints and direction.
How it is calculated
The chosen decision x-star is the member of feasible set X that minimizes objective f(x).
| Symbol | Meaning | Unit |
|---|---|---|
| x | decision vector | decision-specific |
| f | objective function | objective-specific |
| X | feasible decision set | set |
Interpretation
The objective is the scorecard used by the optimizer. Change the scorecard and the selected solution can change even with the same data. Constraints determine which candidates the scorecard is allowed to consider.
Assumptions and conventions
-
The feasible set is defined and nonempty.
-
A minimum is attained, and x-star is selected among minimizers.
-
The objective represents the intended estimand or decision criterion.
-
A minimization or maximization direction is explicit.
The objective retains its own units; x retains decision units.
Limitations and common misreadings
-
An optimizer cannot recover preferences that were omitted from the objective.
-
The best in-sample objective value is not automatically the best out of sample.
References
- Stephen Boyd, Lieven Vandenberghe. (2004). Convex Optimization. Cambridge University Press. Section 4.1, Optimization problems. doi:10.1017/CBO9780511804441. Source

