Constraint annealing: Adaptive ɛ-relaxation on the dual formulation for constrained optimization

The dual formulation recasts a constrained single-objective problem as a multi-objective one by promoting the constraints’ demand-to-capacity ratios into an additional objective spanning the cost–safety trade-off. Enforcing strict feasibility ( d i ≤ 1 ) throughout the search can keep the optimizer from crossing temporarily infeasible regions that act as stepping stones to better designs. We propose constraint annealing : a progressive ɛ -relaxation of the feasibility boundary ( d i ≤ 1 + ɛ ( t ) , ɛ ( t ) → 0 ) that expands the accessible region early and contracts to strict feasibility at termination, inspired by (but not equivalent to) simulated annealing’s cooling schedule. Theoretically, the ɛ -policies form a containment ladder rooted at strict feasibility, and final quality as a function of the relaxation magnitude ɛ max admits, under mere continuity, an interior optimum ɛ max ∗ > 0 ; thus a nonzero relaxation can improve on strict feasibility, chiefly on actively-constrained problems, while any schedule with ɛ → 0 reduces at termination to the strict problem and, under standard global-convergence assumptions, inherits the base optimizer’s convergence. On a matched-budget battery (ten benchmarks, six algorithms, twenty runs), the schedule shape is decisive: the power-law schedule significantly outperforms the linear and exponential ones (Friedman, p < 1 0 − 4 ). It matches strict feasibility overall and improves on it by 14%–55% on actively-constrained problems while tying elsewhere, reaching its final quality up to 20 × faster. Every delivered design is feasible by construction; the only failure mode — an empty feasible front — is rare ( < 1 % for the recommended schedule), confined to the narrowest-feasible problems and shared by strict feasibility.

Authors

Institutions

Publication Details

Journal
Swarm and Evolutionary Computation
Published
2026-11-14
DOI
https://doi.org/10.1016/j.swevo.2026.102615
Primary Topic
Advanced Multi-Objective Optimization Algorithms
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Constraint annealing: Adaptive ɛ-relaxation on the dual formulation for constrained optimization

Jesús A. Torrecilla-Pinero, José Maria Ceballos, Juan A. Gómez-Pulido
Swarm and Evolutionary Computation
Advanced Multi-Objective Optimization Algorithms
article

Constraint annealing: Adaptive ɛ-relaxation on the dual formulation for constrained optimization

Jesús A. Torrecilla-Pinero, José Maria Ceballos, Juan A. Gómez-Pulido
article en

Abstract

The dual formulation recasts a constrained single-objective problem as a multi-objective one by promoting the constraints’ demand-to-capacity ratios into an additional objective spanning the cost–safety trade-off. Enforcing strict feasibility ( d i ≤ 1 ) throughout the search can keep the optimizer from crossing temporarily infeasible regions that act as stepping stones to better designs. We propose constraint annealing : a progressive ɛ -relaxation of the feasibility boundary ( d i ≤ 1 + ɛ ( t ) , ɛ ( t ) → 0 ) that expands the accessible region early and contracts to strict feasibility at termination, inspired by (but not equivalent to) simulated annealing’s cooling schedule. Theoretically, the ɛ -policies form a containment ladder rooted at strict feasibility, and final quality as a function of the relaxation magnitude ɛ max admits, under mere continuity, an interior optimum ɛ max ∗ > 0 ; thus a nonzero relaxation can improve on strict feasibility, chiefly on actively-constrained problems, while any schedule with ɛ → 0 reduces at termination to the strict problem and, under standard global-convergence assumptions, inherits the base optimizer’s convergence. On a matched-budget battery (ten benchmarks, six algorithms, twenty runs), the schedule shape is decisive: the power-law schedule significantly outperforms the linear and exponential ones (Friedman, p < 1 0 − 4 ). It matches strict feasibility overall and improves on it by 14%–55% on actively-constrained problems while tying elsewhere, reaching its final quality up to 20 × faster. Every delivered design is feasible by construction; the only failure mode — an empty feasible front — is rare ( < 1 % for the recommended schedule), confined to the narrowest-feasible problems and shared by strict feasibility.

Swarm and Evolutionary ComputationVol. 111
Universidad de Extremadura (ES)
Ministerio de Ciencia, Innovación y Universidades, Universidad de Extremadura, European Regional Development Fund, Agencia Estatal de Investigación
Advanced Multi-Objective Optimization Algorithms
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon