What question did this study set out to answer?

The central aim is to explore the failure mode of CMA-ES in optimizing under barrier constraints.

April 11, 2026Open Access

Cross-Dimensional Covariance Adaptation Induces Irreversible Drift Toward Deceptive Attractors in Barrier-Constrained Search

Key Points

The central aim is to explore the failure mode of CMA-ES in optimizing under barrier constraints.
Analyzed drift rates of full CMA-ES and sep-CMA-ES in barrier-constrained scenarios.
Conducted population scaling experiments to assess separation of feasible basins.
Evaluated the effectiveness of reactive switching versus preemptive variant selection.
Full CMA-ES exhibited a 73% drift rate towards infeasibility compared to 43% for sep-CMA-ES.
Population scaling showed complete separation of feasible outcomes at lambda=200.
Reactive switching was ineffective; preemptive variant selection was necessary.

Abstract

We report covariance-induced irreversible drift, a failure mode of CMA-ES in barrier-constrained optimization. Full CMA-ES systematically transitions from feasible to infeasible solutions through cross-dimensional covariance adaptation (73% drift rate vs 43% for sep-CMA-ES, Fisher p=0.018). The underlying landscape contains empirically disconnected feasible basins linked by directed, optimizer-dependent transitions. Population scaling experiments show complete separation (100% vs 0%) at lambda=200. Adaptive mitigation via reactive switching fails; preemptive variant selection is required.

Cross-Dimensional Covariance Adaptation Induces Irreversible Drift Toward Deceptive Attractors in Barrier-Constrained Search

Key Points

Abstract

Cite This Study

Also Consider

Also Consider