On Optimality Conditions for Mathematical Programming Problems Based on Strong Subdifferentials

Abstract: We develop refined Karush-Kuhn-Tucker (KKT) and Fritz-John (FJ)-type optimality conditions for nonsmooth, nonconvex mathematical pro-gra-mming problems. We pay special attention in the case that the functional constraint belongs to a specific class of generalized convex functions known as strongly quasiconvex functions. After analyzing a specialized sub-di-ffe-ren-tial, named the strong subdifferential, we compute the normal cone of the supremum function in terms of such subdifferentials, and apply this result to the mathematical programming problem. We illustrate our important results by examples.

• The paper presents refined KKT and Fritz-John optimality conditions based on strong subdifferentials to overcome classical limitations in nonsmooth, nonconvex problems.
• It rigorously develops the properties of strong subdifferentials, ensuring nonemptiness and robust calculus rules for strongly quasiconvex functions.
• The results have broad implications for algorithm design and applications in fields like economics and engineering, providing robust optimality criteria where traditional methods fail.

Optimality Conditions in Nonsmooth Nonconvex Optimization via Strong Subdifferentials

Introduction and Motivation

This work provides a rigorous advancement in the study of optimality conditions for mathematical programming problems characterized by nonsmooth and nonconvex objectives or constraints, with a particular focus on the subclass where the functional constraints exhibit strong quasiconvexity. The existing literature has extensively investigated optimality conditions—such as the Karush-Kuhn-Tucker (KKT) and Fritz-John (FJ) conditions—primarily under the assumptions of convexity or standard quasiconvexity. The major limitation of these classical approaches becomes apparent in the presence of functions with large flat regions, where conventional subdifferential constructions may become empty, impeding the derivation of practical and informative optimality conditions.

To overcome this, the authors introduce refined KKT and FJ-type optimality conditions framed in terms of a specialized nonsmooth analysis tool: the strong subdifferential. This subdifferential is specifically tailored to strongly quasiconvex functions, allowing for the nonemptiness and computational accessibility that is otherwise lacking in the classical limiting or regular subdifferentials in nonsmooth, nonconvex environments.

Theoretical Foundations: Generalized Convexity and Strong Subdifferentials

The background section recapitulates key concepts in variational analysis and generalized convexity, establishing the hierarchy:

• Strong convexity ⇒ Strict convexity ⇒ Convexity
• Strong quasiconvexity ⇒ Strict quasiconvexity ⇒ Quasiconvexity

Strong quasiconvex functions—originally due to Polyak—are central to this discussion because their structural properties ensure the absence of "flat" nonminimal regions, facilitating both the design of optimization algorithms with strong convergence guarantees and the derivation of sharp necessary and sufficient optimality conditions.

The strong subdifferential, denoted $\partial_{\beta,\gamma}^{K} h(\overline{x})$, is defined via a quadratic support-type inequality parameterized by $\beta>0$ and modulus $\gamma\geq0$ over a set $K$, and encodes strong convexity or strong quasiconvexity information not captured by classical constructions. Essential properties rigorously established include:

• Nonemptiness for strongly quasiconvex and lower semicontinuous functions,
• Compactness when restricted to the interior of the domain,
• Calculus rules for supremum and composition of functions,
• Strict inclusions with classical subdifferentials and Greenberg-Pierskalla/quasiconvex subdifferentials under mild conditions.

Characterization of Normal Cones via Strong Subdifferentials

A key technical development of the paper is the description of the normal cone of convex level sets of supremum-type constraint functions in terms of strong subdifferentials. For feasible sets $\Omega = \{x: \sup_j g_j(x) \leq 0\}$, the normal cone $N(\Omega, \bar{x})$ at a feasible point is represented, under convexity and suitable generalized regularity conditions, as a convex combination of the strong subdifferentials of the active $g_j$ at $\bar{x}$, augmented by the horizon subdifferentials when necessary.

Notably, the formula

$$N(\Omega, \bar{x}) = \overline{ \operatorname{co} \left\{\sum_{j \in I(\bar{x}): \mu_j > 0} \mu_j \partial_{\beta_j, \gamma_j}^{K_j} g_j(\bar{x}) + \sum_{j \in I(\bar{x}): \mu_j = 0} \partial^\infty g_j(\bar{x})\right\} }$$

is established, where $I(\bar{x})$ is the set of active constraint indices at $\beta>0$0. This result generalizes several prior works and is proven using advanced tools from variational analysis, including Hahn-Banach-type separation arguments, calculus of subdifferentials, and the geometry of tangent and normal cones.

KKT and FJ-Type Optimality Conditions

The paper delivers a generalized FJ/KKT theory for nonsmooth, nonconvex problems where the constraint structure is strongly quasiconvex. Main results are structured as follows:

• Fritz-John (FJ) Necessary Optimality Conditions: For any local minimizer $\beta>0$1, there exist scalars $\beta>0$2, nonnegative multipliers $\beta>0$3, and subdifferential elements from the strong (and horizon, if needed) subdifferentials, satisfying a generalized stationarity inclusion. The formulation avoids standard constraint qualifications and does not require local Lipschitz continuity.
• Karush-Kuhn-Tucker (KKT) Conditions as Specialization: When $\beta>0$4 and a generalized constraint qualification holds (no nonzero abnormal multiplier), the necessary conditions reduce to a KKT system involving only the limiting subdifferential of $\beta>0$5 and the strong subdifferentials of the active constraints.
• Sufficiency Results: Under strong pseudoconvexity or strong quasiconvexity of the objective and constraints, the derived KKT/FJ-type conditions are also sufficient for optimality. The analysis covers quadratic fractional programming, where the quotient structure is managed using strong subdifferential calculus and ties into fractional programming applications.

Numerical, Structural, and Contradictory Results

Throughout, explicit counterexamples and theoretical claims demonstrate the situations in which the strong subdifferential offers nontrivial and informative conditions, even where classical subdifferentials are empty. For instance, minimizers and non-minimizers can often be separated via the strong subdifferential-based conditions, while older KKT/FJ variants may admit spurious solutions or become vacuous.

A bold assertion in the paper is that their optimality conditions hold under "mild" assumptions, and the precise hypotheses are exhaustively detailed, including regularity, constraint qualification, and convexity criteria for involved sets and functions. Several examples establish the strictness of inclusions between strong, quasiconvex, and Greenberg-Pierskalla subdifferentials.

Implications and Future Research

The framework developed in this work supports both theoreticians and algorithm designers dealing with highly nonconvex, nonsmooth, and non-Lipschitz mathematical programs, especially in situations where generalized convexity notions (strong quasiconvexity) are relevant. The strong subdifferential is a robust analytical instrument for such problems, providing a clear pathway to derive necessary and sufficient conditions for optimality in cases where classical convexity-based subdifferential calculus fails. The applicability spans from economic models with diversified preferences to engineering and machine learning scenarios involving fractional and nonconvex constraints.

Potential future directions, directly hinted at in the conclusion and throughout the technical results, include:

• Algorithmic exploitation of strong subdifferential-based optimality, such as new variants of subgradient or proximal point methods,
• Extension to vector, multiobjective, or variational inequality frameworks,
• Further analysis in infinite-dimensional and nonsmooth settings,
• Investigation of constraint qualifications tailored to strong subdifferentials.

Conclusion

This paper advances the field of nonsmooth, nonconvex optimization by providing refined FJ and KKT-type optimality conditions based on the strong subdifferential, which is especially effective for strongly quasiconvex functions. The results substantially generalize classical subdifferential-based conditions, address the shortcomings associated with nonemptiness and informativeness, and open new lines of research in mathematical optimization theory and methods. The rigorous calculus, precise statements of limitations, and detailed examples provide a robust foundation for both theoretical exploration and future algorithmic developments.