EUVIMEDEUVIMED entdecken
Crossref

Squared-variable parametrizations in inequality-constrained optimization: benign nonconvexity, singular geometry, and second-order transfer

Kan-Lin Hsiung

International Journal of Computing and Optimization · 2026 · 13 (1) · 1-19

DOI: 10.12988/ijco.2026.10017 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Computing and Optimization, 2026. Verfasst von Kan-Lin Hsiung.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Squared-variable parametrizations encode nonnegativity through algebraic identities such as $x=v^{2}$ or $c_i(x)-s_i^{2}=0$, thereby replacing inequality constraints by equalities. Their appeal is clear, but so is the cost: the lift is singular at active constraints, the lifted variables acquire sign symmetry, and the reformulated problem is usually nonconvex even when the original problem is convex. This survey argues that these effects are best understood not as implementation details or fatal pathologies, but as a coherent second-order phenomenon. For simplex constraints, the Hadamard lift to the sphere preserves the KKT/strict-saddle dichotomy, so convex simplex problems inherit a benign nonconvex landscape. For polyhedral feasible sets, the lift produces a singular algebraic variety whose tangent cones and second-order tangent sets support multiplier recovery and optimality certification without additional constraint qualifications. For bound-constrained and general nonlinear programs, exact and approximate transfer theorems link second-order properties of direct square substitution and squared-slack formulations to those of the original problem, with quantitative residual bounds governed by thresholded active-set information. The resulting picture is not that squaring is costless, but that its singularities are structured enough to support a rigorous theory of landscape, certification, and algorithmic transfer.

Zitieren

APA

Hsiung, K. (2026). Squared-variable parametrizations in inequality-constrained optimization: benign nonconvexity, singular geometry, and second-order transfer. International Journal of Computing and Optimization, 13(1), 1-19. https://doi.org/10.12988/ijco.2026.10017

Vancouver

Hsiung K. Squared-variable parametrizations in inequality-constrained optimization: benign nonconvexity, singular geometry, and second-order transfer. International Journal of Computing and Optimization. 2026;13(1):1-19. doi:10.12988/ijco.2026.10017

BibTeX
@article{Hsiung2026,
  author = {Hsiung, Kan-Lin},
  title = {Squared-variable parametrizations in inequality-constrained optimization: benign nonconvexity, singular geometry, and second-order transfer},
  journal = {International Journal of Computing and Optimization},
  year = {2026},
  volume = {13},
  number = {1},
  pages = {1--19},
  doi = {10.12988/ijco.2026.10017}
}
Kostenlos starten →