Revisiting the Constant-Rank Constraint Qualification for Second-Order Cone Programs

Abstract: The constant rank constraint qualification (CRCQ) for second-order cone programs, introduced by Andreani et al. in [Math. Program. 202 (2023), 473 - 513], shares some desirable properties with its classical nonlinear programming counterpart; specifically, it guarantees strong second-order necessary conditions for optimality, and is independent of the Robinson constraint qualification. However, unlike the classical version, this new CRCQ can fail in the linear case, and it is unclear whether CRCQ implies the metric subregularity constraint qualification (MSCQ). The aim of this paper is to examine the CRCQ for second-order cone programs in the linear setting. First, we show that the facial constant rank property, which is a key requirement for the validity of CRCQ, does not always hold in this context. Then, we derive a necessary and sufficient condition for a feasible point to satisfy this property. After that, we establish an easily verifiable characterization of CRCQ. Finally, utilizing this characterization, we prove that CRCQ and MSCQ are equivalent.

• The paper establishes that CRCQ and MSCQ are equivalent in affine SOCPs, offering a fresh perspective on constraint qualifications.
• The paper develops a necessary geometric-algebraic condition for the facial constant rank (FCR) property and provides counterexamples where FCR fails under small perturbations.
• The paper characterizes when the normal image set is closed, a result that underpins second-order optimality conditions and influences algorithmic convergence.

Revisiting the Constant-Rank Constraint Qualification for Second-Order Cone Programs

Introduction and Motivation

The constant rank constraint qualification (CRCQ) is a pivotal concept in nonlinear programming, serving as a foundational tool for establishing stability, convergence, and strong optimality results. The linear independence constraint qualification (LICQ) is often too restrictive, while weaker conditions such as the metric subregularity constraint qualification (MSCQ) are central for various nonsmooth analysis techniques but may lack necessary strength for second-order results. In the context of second-order cone programming (SOCP)—a nonpolyhedral, conic extension of convex programming—recent advances have revisited constraint qualifications, specifically adapting CRCQ to SOCP and assessing its practical and theoretical landscape (2604.00365).

This work critically examines the role and properties of the CRCQ in the context of affine SOCPs. It identifies limitations and obstructions that arise when transferring the intuition and guarantees of CRCQ from the classical nonlinear or polyhedral setting to second-order cone constraints. These findings have significant implications both for variational analysis theory and for the certification of optimality in conic programming.

Facial Constant Rank (FCR) Property in SOCP

A central technical requirement underlying CRCQ for conic programming is the so-called facial constant rank (FCR) property. Unlike the polyhedral setting, where CRCQ (and, hence, FCR) is automatically satisfied in generic linear programs, the nonpolyhedral, specifically second-order cone, setting allows for failure of the FCR—even in simple affine cases.

The authors provide a counterexample that explicitly demonstrates the non-universality of the FCR property at feasible points in linear SOCPs. Formally, the FCR property demands that for every face $F$ of the reducible cone corresponding to the constraint, the dimension of the adjoint image of the orthogonal face $F^\perp$ under the linearized constraints remains constant locally. The counterexample constructs an affine SOCP where, at the origin, this dimension fails to be stable under small perturbations.

A notable result of this work is a necessary and sufficient geometric-algebraic condition for FCR at any feasible point in affine SOCPs. The conditions partition the feasible locus using the position of the constraint image (interior, relative boundary, or zero) and, in the boundary case, whether a certain nondegeneracy (local surjectivity) is satisfied or the reduced constraint trivially vanishes nearby.

Importantly, the FCR fails to be locally preserved—meaning that even if a feasible point exhibits FCR, an infinitesimal perturbation may result in loss of this property. This instability undermines direct transplantation of classical intuitions about constraint regularity into the conic setting.

Closedness of the Normal Image Set $H(\bar{x})$

An equally significant technical point is the closedness of the normal image set $$H(\bar{x}) = \nabla g(\bar{x})^*\left[N_{\mathcal{Q}_m}(g(\bar{x}))\right]$$, which is crucial for characterizations of normal cones, tangent cones, and the validity of second-order results. The analysis leverages structure theory of linear images of cones (notably results from Pataki and Rockafellar) and the geometry of faces in the Lorentz cone.

The authors provide a complete classification: $H(\bar{x})$ is closed if and only if (i) $g(\bar{x})$ is nonzero, (ii) $g(\bar{x}) = 0$ but the image of $A$ intersects the interior of the cone, is trivial, or forms a ray along a boundary direction. Non-closedness arises precisely in pathological cases where the image of $A$ aligns with an extreme direction of the cone without generating the full one-dimensional face.

This structural insight validates and explains previous examples in the literature where the failure of closedness leads to breakdowns in traditional optimality theory.

Verifiable Characterization and Equivalence of CRCQ and MSCQ

By synthesizing the geometric and algebraic characterizations of both FCR and the closedness requirement, the paper provides a fully constructive characterization of CRCQ in linear SOCPs. These conditions are strictly stronger than standard constraint qualifications such as Robinson's constraint qualification (RCQ) and evoke the need to examine the nature of the affine image relative to the cone faces.

A key claim established is the equivalence of CRCQ and MSCQ in the affine second-order cone case. Explicitly, CRCQ holds at a feasible point if and only if the metric subregularity constraint qualification does. The proof utilizes both the verifiable conditions provided for CRCQ and various decompositions of the feasible set in terms of the kernel and image of $A$, linking geometric intuition with metric regularity and properties of projections.

This equivalence deviates from the classical hierarchy of constraint qualifications (where CRCQ properly contains MSCQ), highlighting a structural rigidity of affine SOCPs. In the nonpolyhedral, nonlinear, or reducible settings, such an equivalence does not necessarily obtain, underscoring a unique feature in the affine SOCP case.

Implications and Directions for Future Research

The findings have multiple consequences:

• Optimality theory: The available second-order necessary and sufficient conditions for optimality in SOCPs, which traditionally require CRCQ, are now seen to rest on the more accessible, well-understood MSCQ, provided the problem is affine. This facilitates the application of variational analysis and improves tractability for verification in practical algorithms.
• Algorithmic implications: Since MSCQ is essential for stability and convergence of algorithms in degenerate, nonpolyhedral contexts, the new equivalence sharpens the toolkit for both theoretical convergence and failure analysis.
• Theory of constraint qualifications: The instability of FCR in affine SOCPs and its explicit characterization indicate how nonpolyhedral geometry admits forms of degeneracy that disrupt classical regularity assumptions, encouraging further study on direct generalizations to semidefinite and more complex conic systems.
• Pathological structures: By pinning down exactly when closedness fails, the work gives a pathway for screening and redesigning constraint systems to avoid these degeneracies.

A natural extension is to examine whether analogous equivalences and characterizations hold for affine semidefinite programs and other reducible cones. Systematic identification of practical settings or applications where CRCQ (now MSCQ) holds or fails will further bridge the gap between theoretical advances and use in optimization practice.

Conclusion

This paper provides a comprehensive investigation of CRCQ for affine second-order cone programs, clarifying both its structural limitations and its concrete relationship with MSCQ. Through rigorous characterizations and counterexamples, it is shown that CRCQ and MSCQ are equivalent in the affine SOCP context, diverging from both polyhedral and nonlinear conic settings. These insights inform both theoretical analysis and the practical verification of constraint regularity necessary for advanced conic programming algorithms and optimality conditions.