Quasiconvexity for Continuous Functions

• Quasiconvexity is defined for continuous functions by the convexity of their sublevel sets, offering a generalization of classical convexity.
• This property underpins critical methods in optimization and variational analysis, ensuring existence of maximizers and governing regularity properties.
• Advances in strong quasiconvexity and its preservation under various operations provide insights into uniqueness of minimizers and robust variational techniques.

A real-valued function defined on a convex set is called quasiconvex if its sublevel sets are convex, or equivalently, if it satisfies a weakened form of Jensen's inequality along line segments. Quasiconvexity generalizes convexity and appears naturally in optimization, variational analysis, and the study of lower semicontinuity properties of integral functionals. For continuous functions, quasiconvexity is central both in functional analysis and in the calculus of variations, where regularity, maximization principles, and structure of level sets are governed by this property.

1. Quasiconvexity: Definitions and Characterizations

Let $C$ be a convex subset of a topological vector space $V$. A function $f: C \to [-\infty, \infty)$ is quasiconvex if, for all $x, y \in C$ and all $\lambda \in [0,1]$,

$$f(\lambda x + (1-\lambda)y) \leq \max\{f(x), f(y)\}$$

Equivalently, all sublevel sets $\{x \in C : f(x) \leq t\}$ are convex for each $t \in \mathbb{R}$ (Ball, 2023, Rabier, 2012).

This property is strictly weaker than convexity; every convex function is quasiconvex, but not conversely. The definition admits further generalizations: for instance, strong quasiconvexity introduces a modulus $\sigma>0$ enforcing quantitative descent in the max-inequality,

$$f(\lambda x + (1-\lambda)y) \leq \max\{ f(x), f(y) \} - \frac{\sigma}{2} \lambda(1-\lambda) \|x-y\|^2$$

with $\sigma$-quasiconvexity bridging between ordinary and strong quasiconvexity (Nam et al., 2024).

In infinite-dimensional topological vector spaces, these characterizations extend via the convexity of lower level sets and variation principles based on essential suprema and infima (Rabier, 2012).

2. Continuity Properties and Baire Category Criteria

Quasiconvex functions need not be continuous on their domain. However, on Baire topological vector spaces (comprising Banach and Fréchet spaces), the size and structure of the sublevel sets determine points of continuity. (Rabier, 2012) provides a complete criterion: Let $m = \operatorname{Tessinf}_X f$ be the topological essential infimum. Then:

• The discontinuity set of $f$ is of first category (hence, $f$ is continuous on a residual set) if and only if, for every $a < m$, the set $\{ f \leq a \}$ is nowhere dense whenever it has empty interior, and for every $a > m$, $\{ f \leq a \}$ is nonempty.

A notable dichotomy for the level sets emerges: for all $a \neq m$, $f^{-1}(a)$ either has nonempty interior or is nowhere dense—the structure mimics that of continuous functions except (possibly) at one exceptional level (Rabier, 2012).

Every upper semicontinuous (usc) quasiconvex function is quasicontinuous in Kempisty’s sense: for any open interval $V \subset \mathbb{R}$, the interior of $f^{-1}(V)$ is dense in $f^{-1}(V)$. This follows from the openness of strict superlevel sets and their convexity in the usc, quasiconvex case.

3. Variational Principles and Maximum Theorems

An important extension of classical theorems is the relaxation of convexity to quasiconvexity in maximization results. In the context of a Hausdorff, locally convex topological vector space $V$, for $K \subset V$ nonempty, compact, and convex, if $f: K \to [-\infty, \infty)$ is upper semicontinuous and quasiconvex, then:

• $f$ attains its global maximum on $K$,
• the maximizer set $E = \{x \in K : f(x) = M\}$ is a nonempty, closed, $K$–semi-extremal set,
• and $E$ contains at least one extreme point of $K$ (Ball, 2023).

This "Quasiconvex Maximum Principle" extends Bauer’s maximum principle and is proven by analyzing the semi-extremal structure of the maximizer set, Zorn's lemma, and properties of extreme points. Absent quasiconvexity, continuous functions on compact convex sets may attain their maximum only at interior points, and the correspondence with extreme points can fail.

4. Strong Quasiconvexity, Operations, and Uniqueness

Strong quasiconvexity (with modulus $\sigma > 0$) enforces additional regularity and uniqueness results. Every strongly quasiconvex function is quasiconvex, but not vice versa. Strong quasiconvexity implies:

• Existence and uniqueness of minimizers for lower semicontinuous, strongly quasiconvex functions on reflexive Banach spaces; uniqueness is a consequence of strict quadratic descent at the midpoint (Nam et al., 2024).
• Uniform continuity on bounded sets and a quadratic lower bound implying $2$-supercoercivity.

Operations that preserve $\sigma$-quasiconvexity are scalar multiplication (scaling $\sigma$), monotone composition (with Lipschitz conditions), supremum of families (with $\sigma = \inf_{\alpha} \sigma_{\alpha}$), infimal convolution, and affine transformations (with the modulus scaling depending on the norm of the transformation).

Counterexamples illustrate the fine distinctions: a function can be strongly quasiconvex yet discontinuous, locally but not globally strongly quasiconvex, or the sum of strongly quasiconvex functions can fail to be quasiconvex (Nam et al., 2024).

5. Quasiconvex Envelopes and Regularity Interpolation

A continuum interpolating between convex and quasiconvex functions is constructed via the $A_k$-convexity notion, for $k \in [1, \infty)$. $A_k$-convexity requires that, along any segment, the function lies below the solution of a differential equation

$$v''(t) + (k-1)|v'(t)|^2 = 0, \quad v(0)=u(y), v(1)=u(x).$$

The case $k=1$ recovers convexity, while $k\to\infty$ recovers quasiconvexity. The "interpolating envelope" $E_k[f]$ is the largest $A_k$-convex function not exceeding prescribed boundary data. For each $k$, the envelope solves a degenerate elliptic PDE in the viscosity sense, and as $k \downarrow 1$ or $k \to \infty$, converges to the convex or quasiconvex envelope, respectively (Blanc et al., 2023).

Under a "No V-shaped touching" boundary condition, $E_k[f]$ is $C^1$. At each point, there exists an analog of a supporting hyperplane—a supporting solution to the $A_k$ ODE—illustrating the geometric structure associated to $A_k$-convexity.

6. Lower Semicontinuity and Quasiconvexity in Variational Analysis

In the vector-valued variational context, quasiconvexity provides necessary and sufficient conditions for sequential weak lower semicontinuity of integral functionals. The archetype is Morrey’s quasiconvexity: for $L: \mathbb{R}^{m\times N}\to [0,\infty]$, $L$ is quasiconvex if

$L(F) \leq \int_Y L(F + \nabla \varphi(y))\, dy$

for all compactly supported variations $\varphi$ and any fixed $F$ (Mandallena, 2011, Corbisiero et al., 6 Jan 2026). Extensions include $W^{1, q}$-quasiconvexity for Sobolev-space variations and closed $\mathcal{A}$-$p$-quasiconvexity for sequences subject to constant-rank first-order PDE constraints (Prosinski, 2017).

A localization principle (strengthening classical decomposition lemmas) is essential for sufficiency: Given bounded energy and weak convergence to an affine map, one can construct competitors in the same Sobolev class whose gradients approximate the original gradients and whose energies are equi-integrable. This technical principle underlies the proof that $W^{1, q}$-quasiconvexity plus continuity ensures lower semicontinuity of variational integrals.

On compact Riemannian manifolds, an intrinsic generalization—requiring quasiconvexity with respect to the geometry of the vector bundle of differentials—characterizes the sequential weak* lower semicontinuity of functionals built over $W^{1, \infty}$ spaces (Corbisiero et al., 6 Jan 2026).

7. Geometric and Functional Structure: Level Sets, Maximizers, and Examples

Quasiconvex functions exhibit a categorical dichotomy in the structure of their level sets not present in arbitrary continuous functions. For functions continuous at a residual set, each level set either has nonempty interior or is nowhere dense, up to a single exceptional value (Rabier, 2012). In optimization contexts, quasiconvexity guarantees that maximizers are located at (semi-)extreme points of the domain; this property fails for merely continuous but non-quasiconvex functions (Ball, 2023).

Tables summarizing the relationships of core concepts:

Property	Convex	Quasiconvex	Strongly Quasiconvex
Sublevel sets	Convex	Convex	Convex
Level set dichotomy	Yes	Yes (a.e.)	Yes
Maximum at extreme point	Yes	Yes (under usc)	Yes
Uniqueness of minimizer	Yes	No	Yes (if lsc)
Uniform continuity (bounded)	Yes	No	Yes

These structural and variational features make quasiconvexity central in optimization, nonlinear analysis, and mathematical modeling where generalizations of convexity are essential. For continuous functions specifically, the interplay between regularity, geometric properties, and variational consequences is sharp and well-understood in light of the developments cited above.

---

Referenced works: (Ball, 2023, Rabier, 2012, Nam et al., 2024, Blanc et al., 2023, Prosinski, 2017, Mandallena, 2011, Corbisiero et al., 6 Jan 2026).