DCT Lexicon · Optimization
Genetic search (NSGA-II)
An evolutionary search that keeps a gallery of plans no other proposed plan dominates; it approximates a Pareto frontier when plans cannot be enumerated, and its results are statistical, not certified.
Drafted for the author's approval from Vol. II, Section 12.4 (Computational Implications), Proposition 12.2(iv) and Table 12.2. The NSGA-II algorithm itself (Deb et al.) is cited, not defined, by the book
In depth
In the book
Evolutionary methods, NSGA-II [8] and MOEA/D [31] among them, earn their keep when plans cannot be enumerated and each evaluation is a simulation; their archives are galleries, to be checked by bounds before the board relies on them.
Vol. II, Section 12.4, Computational Implications, verbatim
Formal statement
Evolutionary search. On a finite set of plans, if every plan can be proposed in every generation and an archive keeps every plan no other proposed plan dominates, the archive equals the efficient set after finitely many generations, almost surely.
Vol. II, Proposition 12.2(iv) (Computing Frontiers), verbatim
Guarantee
Evolutionary search finds a gallery of plans; its guarantee is statistical, not certified (Proposition 12.2); watch for archives that forget.
Vol. II, Table 12.2, verbatim
Common misreading
A common mistake is to trust an evolutionary gallery without checking it by bounds, or to treat an appetite as a fact rather than a point chosen on a frontier.
Vol. II, LOS 12.5, verbatim
How AXIOM uses it
The DCT Simulator's genetic search trades value against the chance of success over integer and non-linear levers, which the integer program cannot hold together. It runs on 5,000 preview paths with an archive that forgets nothing; its finalists are certified on 100,000 fresh paths and checked against the integer program.
Drafted for the author's approval, from AXIOM's method passport "Genetic search (NSGA-II)" (D-098; in flight on 29 October 2026, not yet on main)
In the book: Vol. II, Ch. 12: Multi-Objective Enterprise Optimization
Related: Pareto Frontiers · Mixed-integer programming (MILP) · Optimal (with solver evidence)
V2.Prop.12.2 · V2.Tab.12.2 · V2.Def.12.1 · GD.09


