Mitteoptimaalsust

Mitteoptimaalsust is a term used in optimization theory to describe the property of a solution, model, or decision that is not optimal with respect to a defined objective function under given constraints. It denotes a deviation from the best possible outcome within the feasible set.

Origin and usage: The word is formed from Estonian components “mitte” meaning not and “optimaalsust” meaning optimality. In Estonian-language texts the term is sometimes used as a direct translation of non-optimality or suboptimality, though many authors prefer the standard terms suboptimality or non-optimality.

Formal framing: Let f be an objective to minimize (or maximize) over a feasible set X. Let x* ∈ X be a global optimum, i.e., f(x*) ≤ f(x) for all x ∈ X in the minimization case (or f(x*) ≥ f(x) for all x in maximization). A point x ∈ X exhibits mitteoptimaalsust if f(x) > f(x*) for minimization, or f(x) < f(x*) for maximization. The suboptimality gap is f(x) − f(x*). Local non-optimality may occur when x is not better than nearby feasible points, even if global optima exist elsewhere.

Examples and implications: In a linear program minimizing cost, a feasible solution with cost higher than the optimum is mitteoptimaalsust. In machine learning, a model with higher training loss than the minimum achievable (given data and regularization) is suboptimal. In practice, decisions are often traded off against computation time, robustness, or data availability, so mitteoptimaalsust is not always undesirable.

See also: Suboptimality, optimality, local minimum, global optimum.