🔢 mathematics

An Equivalence Result on the Order of Differentiability in Frobenius' Theorem

This paper resolves an asymmetry in the differentiability of solutions versus integral manifolds within Frobenius' Theorem by demonstrating that integral manifolds possess one degree higher regularity than the underlying system, while also providing counterexamples and characterizing conditions for quasi-convex solutions in optimization contexts.

Original authors: Yuhki Hosoya

Published 2026-03-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yuhki Hosoya

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a cartographer trying to draw a map of a mysterious, foggy island. You have a compass (let's call it $g$ ) that always points in a specific direction at every spot on the island. Your goal is to draw contour lines (let's call them $u$ ) that represent the "height" of the land.

In a perfect world, if your compass is smooth and perfect, the rules of geometry (a famous theorem by Frobenius) say: "If your compass directions don't twist in a weird way, you can always draw smooth, perfect contour lines."

But this paper asks a very specific, tricky question: What happens if your compass is a bit rough or jagged?

The Core Problem: The "Rough Compass" Paradox

The author, Yuhki Hosoya, investigates what happens when the compass ( $g$ ) isn't perfectly smooth (mathematically, it's not $C^\infty$ or even $C^2$ ).

Here is the weird asymmetry he found:

The Compass ( $g$ ): If the compass is "locally Lipschitz" (a fancy way of saying it's not too jagged, it has a limit to how fast it can change), the map you draw ( $u$ ) might only be "locally Lipschitz" too. It might have a few kinks.
The Contour Lines (The Level Sets): However, the shape of the contour lines themselves (the actual lines you draw on the paper) must be perfectly smooth ( $C^1$ ).

The Analogy:
Imagine you are walking through a forest with a slightly shaky compass. You are trying to trace a path that is always perpendicular to the compass needle.

The path you walk (the solution $u$ ) might be a bit bumpy because your compass is shaky.
But the trail markers (the level sets) you leave behind must form a perfectly smooth, continuous line. You can't have a jagged, broken trail marker, even if your walking path was a bit wobbly.

The paper proves that this "rough path, smooth trail" relationship is the only way the math works. If you try to force the path to be perfectly smooth when the compass is only "okay," you run into a contradiction (a mathematical impossibility).

The "Smoothness Gap"

The paper highlights a "smoothness gap."

If your compass is $C^k$ (smooth to the $k$ -th degree), your path ( $u$ ) will be $C^k$ .
But the trail markers (the level sets) will be one degree smoother ( $C^{k+1}$ ).

Creative Metaphor:
Think of a sculptor (the compass) chipping away at a block of stone to reveal a statue (the contour lines).

If the sculptor's chisel is slightly dull (less smooth), the action of chipping (the path) is rough.
But the surface of the statue being revealed is naturally smoother than the chisel's action. The paper proves that the "surface" of the solution is always one step smoother than the "force" driving it.

The Optimization Puzzle (Finding the Bottom of the Valley)

The second half of the paper tackles a practical problem: Optimization.
Imagine you want to find the lowest point in a valley (the minimum of a function) while staying within certain boundaries (like fences).

Standard Math: Usually, we need the valley to be "convex" (bowl-shaped) to easily find the bottom using standard tools (like the KKT conditions).
The Problem: Sometimes, the valley isn't a perfect bowl, but it's "quasi-convex" (it doesn't have any weird hills inside the valley, even if the walls aren't perfectly curved).
The Paper's Solution: Hosoya shows that even if the valley isn't a perfect bowl, as long as the "compass" (the gradient) follows the rules he discovered, you can still find the lowest point using the same standard tools.

Analogy:
Imagine you are blindfolded and trying to find the bottom of a bowl.

If the bowl is perfectly round (convex), you just walk downhill. Easy.
If the bowl is weirdly shaped but still has no "false bottoms" or "hills" inside it (quasi-convex), you might think you're stuck.
Hosoya says: "Don't worry! As long as the slope of the ground follows the rules of our 'rough compass,' you can still use the same 'walk downhill' strategy to find the true bottom."

Why This Matters

It fixes a hole in the theory: For a long time, mathematicians assumed everything had to be perfectly smooth. This paper says, "No, we can handle rougher, more realistic situations, but we have to be careful about which part is smooth and which isn't."
It helps economists and engineers: Many real-world problems (like pricing goods or designing structures) involve functions that aren't perfectly smooth. This paper gives them a rigorous way to solve optimization problems in those messy, real-world scenarios without needing to pretend the world is perfectly smooth.

Summary in One Sentence

This paper proves that when you have a slightly imperfect "compass" guiding you, the path you take might be a bit rough, but the boundaries you trace will always be smoother than the compass itself, allowing you to solve complex "find the best spot" problems even in imperfect, bumpy environments.

1. Problem Statement

The paper addresses a fundamental asymmetry in the regularity (differentiability) of solutions to total differential equations of the form:
$\nabla u(x) = \lambda(x)g(x)$
where $U \subset \mathbb{R}^n$ is an open set and $g: U \to \mathbb{R}^n \setminus \{0\}$ is a given vector field.

Classical Context: Under the assumption that $g$ is $C^\infty$ , Frobenius' Theorem states that the existence of a local solution $u$ (and $\lambda$ ) is equivalent to Jacobi's integrability condition (a specific partial differential equation involving the derivatives of $g$ ).
The Gap: When the smoothness assumption on $g$ $g$ is relaxed (e.g., $g$ $g$ is only $C^k$ $C^{k}$ or locally Lipschitz), a discrepancy arises.
- If $g$ is $C^k$ , one can guarantee the existence of a $C^k$ solution $u$ .
- However, the integral manifolds (the level sets of $u$ , defined as $u^{-1}(c)$ ) are actually $C^{k+1}$ .
- Conversely, if $g$ is only locally Lipschitz, the solution $u$ is only guaranteed to be locally Lipschitz, yet its level sets are $C^1$ manifolds.
Specific Counterexample: The paper references Debreu's counterexample where $g$ is $C^1$ , satisfying the integrability condition, yet no $C^2$ solution $u$ exists. This highlights that the solution $u$ cannot always match the differentiability of the integral manifolds.

The paper aims to resolve this asymmetry by characterizing the exact differentiability requirements for the solution $u$ versus the geometry of its level sets (integral manifolds) under varying smoothness assumptions on $g$ .

2. Methodology

The author employs a combination of differential geometry, ordinary differential equations (ODEs), and convex analysis.

Normal Solutions: The paper introduces the concept of a "normal solution" $u$ $u$ . This is a locally Lipschitz function where:
1. Level sets are $(n-1)$ -dimensional $C^1$ manifolds.
2. $g$ is a normal vector field to these level sets.
3. $u$ has no local extrema.
Reduction to ODEs: The core technique involves transforming the partial differential equation (PDE) $\nabla u = \lambda g$ into a system of ODEs. By assuming $g_n \neq 0$ , the author defines $f_i = -g_i/g_n$ and constructs a parametrized ODE:
$\dot{c}(t) = f((1-t)\tilde{x}^* + t\tilde{x}, c(t)) \cdot (\tilde{x} - \tilde{x}^*)$
The solution to this ODE is used to construct the level sets of $u$ via the implicit function theorem.
Rademacher's Theorem: Since $g$ is assumed to be locally Lipschitz, the author utilizes Rademacher's theorem (which states that Lipschitz functions are differentiable almost everywhere) to handle the integrability conditions almost everywhere.
Convex Analysis for Optimization: In the application section, the paper uses the Separating Hyperplane Theorem and properties of quasi-convex functions to derive optimality conditions (KKT conditions) for minimization problems where the objective function is a solution to the differential equation.

3. Key Contributions and Results

A. Equivalence Theorem for Differentiability (The Main Result)

Theorem 1 establishes a precise equivalence between the integrability of $g$ and the existence of solutions with specific regularity:

Locally Lipschitz Case: If $g$ is locally Lipschitz, $g$ satisfies Jacobi's integrability condition (almost everywhere) if and only if for every $x^*$ , there exists a locally Lipschitz solution $u$ such that every level set of $u$ is a $C^1$ manifold.
- Significance: This resolves the asymmetry. Even if $u$ is only Lipschitz, the geometry of its level sets is strictly $C^1$ .
$C^k$ Case: If $g$ is $C^k$ ( $k \geq 1$ ), $g$ satisfies Jacobi's condition (everywhere) if and only if for every $x^*$ , there exists a $C^k$ solution $u$ such that every level set of $u$ is a $C^{k+1}$ manifold.
- Significance: This proves that the integral manifolds always possess one degree of higher differentiability than the solution function $u$ itself.

B. Uniqueness of Level Sets

Proposition 2 and Corollary 1 demonstrate that if two normal solutions $u_1$ and $u_2$ exist for the same $g$ , they are related by a monotone transformation ( $u_2 = \phi \circ u_1$ ). Consequently, they share the exact same level sets (integral manifolds). This implies that the geometric structure (the foliation) is unique, even if the scalar potential function $u$ is not.

C. Optimization and Quasi-Convexity

The paper applies these results to optimization problems where the objective function $u$ is a solution to the differential equation.

Theorem 2: Establishes that if $u$ is quasi-convex, the Karush-Kuhn-Tucker (KKT) conditions are necessary and sufficient for a point to be a minimizer, even without assuming $u$ is convex. This is significant because solutions to (1) are often quasi-convex but not convex.
Theorem 3 & Proposition 3: Provides necessary and sufficient conditions on the vector field $g$ $g$ that ensure any regular solution $u$ $u$ is quasi-convex (or strictly quasi-convex).
- Condition (i): A geometric condition on $g$ involving dot products.
- Condition (ii): A condition on the limit superior of directional derivatives of $g$ .
- These conditions allow economists and mathematicians to verify quasi-convexity directly from the vector field $g$ without solving for $u$ .

D. Counterexamples

The paper includes rigorous counterexamples to clarify the boundaries of the theory:

Debreu's Example: Re-analyzed to show that while a $C^1$ solution exists, no $C^2$ solution exists for a specific $C^1$ vector field, confirming the $C^k$ vs $C^{k+1}$ gap.
Arrow-Enthoven Example: Demonstrates a case where all solutions are quasi-convex, but no convex solution exists. This justifies the paper's focus on quasi-convexity rather than convexity.
Katzner's Example: Shows a case where solutions are strictly quasi-convex, but the stronger condition (II) in Proposition 3 is violated, proving that the conditions are not all equivalent.

4. Significance

Mathematical Rigor: The paper refines Frobenius' Theorem for non-smooth settings, which is crucial for applications in economics and control theory where data or utility functions are rarely $C^\infty$ .
Resolution of Asymmetry: It formally characterizes the "loss" of differentiability in the potential function $u$ versus the "gain" in the smoothness of the integral manifolds.
Economic Application: The results are highly relevant to consumer theory and general equilibrium. In economics, utility functions are often only required to be quasi-convex (representing convex preferences) rather than convex. The paper provides the theoretical foundation for applying KKT conditions to such non-convex but quasi-convex utility functions derived from differential equations.
Robustness: By relying on locally Lipschitz assumptions and Rademacher's theorem, the results are applicable to a broader class of problems than classical smooth manifold theory.

In summary, Hosoya provides a comprehensive framework linking the regularity of a vector field to the regularity of its integral manifolds and the potential functions defining them, while extending these insights to optimization problems involving quasi-convex objectives.