Extending quasiconvex functions from uniformly convex sets

This paper investigates the extension of Lipschitz quasiconvex functions from closed convex sets with nonempty interior in finite-dimensional normed spaces, demonstrating that while Lipschitz extensions generally fail unlike in the convex case, the existence of uniformly continuous or continuous extensions is determined by specific geometric properties of the domain set.

The Gist

Here is an explanation of the paper "Extendability of Quasiconvex Functions in Finite-Dimensional Normed Spaces" using simple language, everyday analogies, and metaphors.

The Big Picture: The "Fence" Problem

Imagine you have a piece of land, let's call it $C$. This land is a specific shape (a convex shape, meaning if you draw a line between any two points on the land, the whole line stays inside the land).

On this land, you have a landscape (a function, $f$). This landscape has hills and valleys. The specific rule for this landscape is that it is Quasiconvex.

• What does that mean? Imagine you are walking down a hill. If you are at a certain height (say, 100 meters), the area where you are below 100 meters is a single, connected chunk of land. It doesn't have weird islands or disconnected pieces. It's a "bowl" shape, though not necessarily a perfect mathematical bowl.

The Problem:
You want to expand your map. You want to take the landscape you built on your private land ($C$) and extend it to cover the entire world ($X$), which is a much bigger space.

The big question the authors ask is: Can we always extend this landscape to the whole world without breaking the rules?

Specifically, they ask three versions of this question:

1. The Smoothness Test (Lipschitz): Can we extend it so the hills don't get suddenly super steep? (No sudden cliffs).
2. The Consistency Test (Uniformly Continuous): Can we extend it so that if two points are close together on the map, their heights are also close together everywhere?
3. The Basic Test (Continuous): Can we extend it so there are no sudden jumps or gaps in the terrain?

---

The Surprising Twist: It's Not Like a Rubber Sheet

In the world of Convex functions (perfectly smooth, bowl-shaped landscapes), the answer is easy. There is a famous rule (McShane's extension) that says: Yes, you can always stretch a convex landscape to cover the whole world without breaking it. It's like stretching a rubber sheet; it just works.

But for Quasiconvex functions, the authors say: "Not so fast!"

They discovered that unlike the perfect rubber sheet, quasiconvex landscapes are fragile. Whether you can stretch them to cover the whole world depends entirely on the shape of your land ($C$).

The Three Main Discoveries

1. The "Cliff" Problem (Lipschitz Extensions)

The Finding: You generally cannot extend a quasiconvex landscape to the whole world while keeping it "smooth" (Lipschitz), unless your land is very simple (like a flat line).

The Analogy: Imagine your land is a valley that gets narrower and narrower as it goes out, like a funnel. If you try to extend the map to the whole world, you might be forced to build a "cliff" (a sudden, infinite drop) just outside your property line to keep the rules of the landscape intact.

• The Result: If your land has "asymptotic directions" (it stretches out infinitely like a funnel or a wedge) or if the edges of your land have flat spots (not perfectly round), you will inevitably hit a wall. You cannot extend the map smoothly.

2. The "Shape" Matters (Continuous Extensions)

The Finding: If you just want to extend the map without sudden jumps (Continuous), the shape of your land is the key.

• Good Shape: Your land must be Rotund (perfectly round, no flat edges) and it must not stretch out to infinity in a way that creates "asymptotic directions" (it must be "closed" in a specific geometric way).
• Bad Shape: If your land has a flat edge (like a square or a rectangle) or if it stretches out infinitely like a long corridor, you cannot extend the map continuously.

The Metaphor: Think of your land as a cookie cutter.

• If the cutter is a perfect circle (Rotund) and it's a finite size, you can smoothly fill in the rest of the dough.
• If the cutter is a square (flat edges) or an infinite strip, the "dough" (the extension) will crack or tear when you try to fill the rest of the table.

3. The "No-Go" Zones (Uniformly Continuous)

The Finding: This is the strictest test. To extend the map so that the "closeness" of points is preserved everywhere (Uniformly Continuous), your land must be Bounded (finite size) and Perfectly Round.

• If your land is infinite (even if it's round), you can't do it.
• If your land is finite but has flat edges, you can't do it.

---

The "Secret Sauce": Geometry is Destiny

The authors spent the paper proving that geometry dictates possibility.

• If your land is a perfect, finite ball: You are safe! You can extend the landscape in almost any way you want (Continuous, Uniformly Continuous, etc.).
• If your land is a square, a rectangle, or an infinite strip: You are in trouble. The math says you cannot extend the landscape without breaking the rules (creating cliffs or jumps).

Why Does This Matter?

You might ask, "Who cares about extending math landscapes?"

These functions are used in Economics (modeling consumer choices), Engineering (optimizing designs), and Computer Science (machine learning).

• In these fields, we often have data on a specific region and want to predict what happens outside that region.
• This paper tells us: "Be careful!" You can't just assume your model will work everywhere. If your data comes from a region with a weird shape (like a long corridor or a flat-sided box), your predictions might break down completely when you try to apply them to the real world.

Summary in One Sentence

While perfect, bowl-shaped landscapes can always be stretched to cover the whole world, quasiconvex landscapes are picky: they can only be extended smoothly if the land they sit on is perfectly round and finite; otherwise, the math breaks, and you can't make the map work.

Technical Summary

Here is a detailed technical summary of the paper "Extendability of Quasiconvex Functions in Finite-Dimensional Normed Spaces" by Carlo Alberto De Bernardi and Libor Veselý.

1. Problem Statement

The paper investigates the extension problem for quasiconvex (QC) functions defined on a closed convex set $C$ within a finite-dimensional normed space $X$ (where $\dim X \ge 2$).

• Context: For convex functions, the McShane extension theorem guarantees that any $L$-Lipschitz convex function on a convex set $C$ can be extended to an $L$-Lipschitz convex function on the entire space $X$.
• The Gap: It is natural to ask if a similar result holds for quasiconvex functions (functions whose sub-level sets are convex).
• Core Question: Under what geometric conditions on the convex set $C$ can a QC function with specific regularity properties (Lipschitz, uniformly continuous, continuous, or upper semicontinuous) defined on $C$ be extended to a QC function on $X$ with the same or weaker regularity?

The authors aim to characterize the extendability of QC functions in terms of the geometric properties of the domain $C$ (specifically boundedness, rotundity/strict convexity, and the existence of asymptotic directions).

2. Methodology

The authors employ a combination of constructive counterexamples and geometric analysis:

1. Counterexample Construction:

   • They construct specific Lipschitz QC functions on various types of convex bodies (unbounded sets, sets with non-strictly convex boundaries) that cannot be extended to the whole space while preserving regularity.
   • Techniques involve defining functions via sub-level sets that "escape" to infinity or create discontinuities upon extension, often utilizing the Hahn-Banach theorem to separate points and constructing sequences of nested convex sets.

2. Geometric Characterization of Domains:

   • Asymptotic Directions: They analyze sets with and without asymptotic directions (directions in which the set extends to infinity without "closing off").
   • Rotundity: They distinguish between general convex sets, rotund (strictly convex) sets, and uniformly rotund sets.
   • Extension Method: They utilize a specific extension technique involving the cone generated by a point and the set, and the intersection of supporting half-spaces (denoted as $e(B)$). This method allows them to build extensions for continuous functions on sets without asymptotic directions.

3. Reduction to 2D:

   • Since the problem is finite-dimensional, many proofs reduce the general $d$-dimensional case to the 2-dimensional Euclidean plane ($\mathbb{R}^2$) using linear projections and quotient spaces, leveraging the fact that any 2D subspace is complemented.

3. Key Contributions and Results

A. Negative Results (Non-Extendability)

The paper establishes that, unlike convex functions, QC functions generally cannot be extended with strong regularity properties unless the domain satisfies strict geometric conditions.

• Lipschitz Extensions:

  • Theorem 5.3: For any convex body $C$ in a normed space of dimension $\ge 2$ (provided $C$ is not affine), there exists a Lipschitz QC function on $C$ that admits no Lipschitz QC extension to $X$.
  • Implication: The "quasiconvex version" of the McShane extension theorem fails completely for Lipschitz functions.
  • Specific Cases:
    • If $C$ is unbounded and admits an asymptotic direction, no QC extension (of any kind) exists for some Lipschitz functions (Theorem 3.1).
    • If $C$ is not rotund (contains a line segment on its boundary), no continuous QC extension exists for some Lipschitz functions (Theorem 3.2).

• Uniformly Continuous Extensions:

  • Theorem 4.1: Even if $C$ is rotund (strictly convex), if it is unbounded and has no asymptotic directions, there exists a Lipschitz QC function that admits no uniformly continuous QC extension. This shows that boundedness is a necessary condition for uniform continuity preservation.

B. Positive Results (Extendability)

The paper provides exact characterizations for when extensions are possible, depending on the desired regularity of the extension.

• Continuous Extensions:

  • Theorem 7.4 & 9.6: A continuous QC function on $C$ admits a continuous QC extension to $X$ if and only if $C$ is rotund and has no asymptotic directions.
  • Note: If $C$ is bounded and rotund, it is automatically uniformly rotund, and extensions exist (recovering previous results). The novelty here is handling unbounded rotund sets without asymptotic directions.

• Upper Semicontinuous (USC) Extensions:

  • Theorem 8.2 & 9.7: A continuous QC function on $C$ admits an upper semicontinuous QC extension to $X$ if and only if $C$ has no asymptotic directions.
  • Note: If $C$ has asymptotic directions, even USC extensions may fail (Proposition 8.3 provides a counterexample on a rectangle).

• Trivial Cases:

  • If $C$ is affine or at most 1-dimensional, extensions preserving Lipschitz/continuity properties always exist (Corollary 9.2).

C. Summary of Characterizations (Theorems 9.5–9.7)

For a finite-dimensional, non-affine, closed convex set $C$ ($\dim \ge 2$):

Desired Extension Property	Necessary and Sufficient Condition on $C$
Lipschitz $\to$ Lipschitz	Never (unless $C$ is affine or 1D).
Lipschitz $\to$ Uniformly Continuous	$C$ is bounded and rotund.
Lipschitz $\to$ Continuous	$C$ is rotund and has no asymptotic directions.
Lipschitz $\to$ Upper Semicontinuous	$C$ has no asymptotic directions.

4. Significance

1. Fundamental Distinction: The paper rigorously demonstrates that the extension theory for quasiconvex functions is fundamentally different from that of convex functions. The "quasiconvex version" of the McShane theorem fails in the strongest sense (Lipschitz preservation is impossible for general convex bodies).
2. Geometric Insight: It establishes a precise link between the analytic properties of function extensions and the geometric properties of the domain (specifically the interplay between strict convexity/rotundity and the behavior at infinity/asymptotic directions).
3. Completeness: The authors provide a complete classification for finite-dimensional spaces, resolving open questions regarding the necessity of the "uniformly rotund" property for continuous extensions and clarifying the role of asymptotic directions.
4. Methodological Advance: The construction of the extension operator $e(B)$ for unbounded sets without asymptotic directions provides a new tool for handling QC functions in non-compact settings.

In conclusion, the paper concludes that while one cannot generally extend QC functions with strong regularity (Lipschitz), one can achieve continuous or upper semicontinuous extensions if the domain is sufficiently "round" (rotund) and does not "open up" to infinity in a way that creates asymptotic directions.