Weak derivatives and metric differentiability almost everywhere

Basic Information

• Paper ID: 2511.02520
• Title: Weak derivatives and metric differentiability almost everywhere
• Author: Nikita Evseev (Okinawa Institute of Science and Technology Graduate University)
• Classification: math.FA (Functional Analysis), math.MG (Metric Geometry)
• Submission Date: November 4, 2025 to arXiv
• Paper Link: https://arxiv.org/abs/2511.02520
• Funding Information: JSPS Grant-in-Aid for Scientific Research No. 25K07038

Abstract

This paper investigates metric differentiability of Lipschitz continuous mappings from Euclidean domains to metric spaces. It is known that such mappings are metric differentiable almost everywhere. When the target space is the dual of a separable Banach space, the metric differential has a linear counterpart—the weak* differential. However, for arbitrary metric spaces or Banach spaces, Lipschitz mappings are not necessarily weak* differentiable. This paper introduces a method based on the concept of "weak weak* derivatives" to provide a linear representation of metric differentials, making them computable as norms of associated linear operators.

Research Background and Motivation

Core Problems

1. Limitations of Classical Results: Kirchheim's (1994) metric Rademacher theorem states that Lipschitz mappings $f: \mathbb{R}^n \to X$ (where $X$ is a metric space) are metric differentiable almost everywhere, meaning there exists a seminorm $\text{md}(f,x)$ such that $\lim_{y\to x} \frac{d(f(y), f(x)) - \text{md}(f,x)(y-x)}{|y-x|} = 0$
2. Need for Linear Representation: Ambrosio and Kirchheim (2000) proved through isometric embedding into the dual $V^*$ of a separable Banach space that when $f: \mathbb{R}^n \to V^*$, the metric differential can be represented as the norm of a weak* differential: $\text{md}(f,x)(\nu) = \|\text{wd}f(x) \cdot \nu\|_{V^*}$
3. Key Challenge: What happens when the target space $X$ itself has linear structure but is not the dual of a separable space? In this case, the mapping is not guaranteed to be weak* differentiable. Although formula (2) can be applied through isometric embedding $\iota: X \to V^*$, the embedding does not respect the original linear structure.

Research Motivation

To find a differential concept suitable for formula (1) that applies not only to the weak* differentiable case but also to more general settings, providing an intrinsic linear representation for metric differentials.

Core Contributions

1. Introduction of Weak Weak Derivative Framework*: Proposes a derivative concept $\nabla^{\circ}f$ applicable to mappings with metric space values, taking values in $L(\mathbb{R}^n; (\text{Lip}_{z_0}(X))^*)$
2. Establishment of Linear Representation Theorem (Theorem 1.1): Proves that for Lipschitz continuous mappings $f: \Omega \to X$:
   • There exists an operator $\nabla^{\circ}f(x)$ satisfying $\langle \phi, \nabla^{\circ}f(x) \cdot \nu \rangle = \nabla\phi \circ f(x) \cdot \nu$
   • There exists a seminorm $\rho$ such that $\text{md}(f,x)(\nu) = \rho(\nabla^{\circ}f(x) \cdot \nu)$ holds almost everywhere
3. Establishment of Norm Representation of Metric Derivative (Theorem 3.5): For absolutely continuous mappings, proves $\|\partial^{\circ}_\nu f(x)\|_{(\text{Lip}_{z_0}(X))^*} = m\partial_\nu f(x)$
4. Sobolev Space Characterization (Theorem 3.7): Characterizes metric-valued Sobolev spaces $W^{1,p}(\Omega; X)$ using weak weak* derivatives
5. Metric Differentiability in Sobolev Topology (Theorem 4.4): Proves that mappings in $W^{1,p}(\Omega; X)$ are metric differentiable almost everywhere in the $W^{1,p}$ topological sense

Detailed Methodology

Core Concept: Weak Weak* Derivatives

Definition (Lemma 3.1)

Let $f: \Omega \to X$ satisfy for all $\phi \in \text{Lip}_{z_0}(X)$: $\limsup_{h\to 0} \frac{|\phi \circ f(x+h\nu) - \phi \circ f(x)|}{|h|} \leq C \text{Lip}(\phi)$

Then there exists a continuous linear functional $\partial^{\circ}_\nu f(x) \in (\text{Lip}_{z_0}(X))^*$ such that:

• When the directional derivative exists, $\langle \phi, \partial^{\circ}_\nu f(x) \rangle = \frac{\partial \phi \circ f}{\partial \nu}(x)$
• $\|\partial^{\circ}_\nu f(x)\|_{(\text{Lip}_{z_0}(X))^*} \leq m\partial_\nu f(x)$

Construction Method: Using Banach limits to define $\langle \phi, \partial^{\circ}_\nu f(x) \rangle := \varphi\text{-}\lim_{n\to\infty} \frac{\phi(f(x+\delta_n\nu)) - \phi(f(x))}{\delta_n}$

Key Technical Tools

1. Essential Norming Sequence (Definition 3.2): For a separable subset $E \subset X$, construct $\{\phi_k\} \subset \text{Lip}_{z_0}(X)$ satisfying $d(x,y) = \sup_k |\phi_k(x) - \phi_k(y)|$
2. Integral Representation of Metric Derivative (Theorem 3.4): If $\phi_k \circ f$ is absolutely continuous on line segments parallel to $\nu$, and $G(x) := \sup_k \left|\frac{\partial \phi_k \circ f}{\partial \nu}(x)\right| \in L^1_{\text{loc}}(\Omega)$ then $m\partial_\nu \tilde{f}(x) = G(x)$ almost everywhere

Proof Strategy for Main Theorems

Proof of Theorem 3.5

Objective: Prove $\|\partial^{\circ}_\nu f(x)\|_{(\text{Lip}_{z_0}(X))^*} = m\partial_\nu f(x)$

Steps:

1. From Lemma 3.1 obtain the inequality $\|\partial^{\circ}_\nu f(x)\|_{(\text{Lip}_{z_0}(X))^*} \leq m\partial_\nu f(x)$
2. For the separable image $f(\Omega \setminus \Sigma)$, construct an essential norming sequence $\{\phi_k\}$
3. By Theorem 3.4 and Lemma 3.1: $m\partial_\nu f(x) \leq \sup_k \left|\frac{\partial \phi_k \circ f}{\partial \nu}(x)\right| = \sup_k |\langle \phi_k, \partial^{\circ}_\nu f(x) \rangle| \leq \|\partial^{\circ}_\nu f(x)\|_{(\text{Lip}_{z_0}(X))^*}$

Lemma 4.1 (Linear Representation of Metric Differential)

Core Conclusion: There exists a seminorm $\rho$ such that $\text{md}(f,x)(\nu) = \rho(\nabla^{\circ}f(x) \cdot \nu)$

Proof Key Points:

1. For a countable dense set $H \subset \mathbb{R}^n$ and essential norming sequence $\{\phi_k\}$, prove $\sup_k \left|\frac{\partial \phi_k \circ f}{\partial \nu}(x)\right| = \|\partial^{\circ}_\nu f(x)\|_{(\text{Lip}_{z_0}(X))^*} = \text{md}(f,x)(\nu)$
2. Extend to all $\nu \in \mathbb{R}^n$ by density argument
3. Use Lipschitz continuity and differentiability of $\phi_k \circ f$: $\langle \phi_k, \nabla^{\circ}f(x) \cdot \nu \rangle = \sum_{j=1}^n \langle \phi_k, \partial^{\circ}_j f(x) \rangle \nu_j$
4. Define seminorm $\rho(w) = \sup_k |\langle \phi_k, w \rangle|$

Experimental Setup

This is a pure theoretical mathematics paper with no numerical experiments. The theoretical framework is established entirely through rigorous mathematical proofs.

Theoretical Verification Methods

1. Counterexample Construction: The paper discusses why general Banach space-valued mappings do not guarantee weak* differentiability
2. Connection with Existing Theory:
   • Comparison with the differential concept of Gigli-Pasqualetto-Soultanis (Section 6)
   • Connection with Reshetnyak's Sobolev space theory (Section 2.3)

Experimental Results

Main Theoretical Results

1. Sobolev Space Characterization (Theorem 3.7)

Conclusion: $f \in W^{1,p}(\Omega; X)$ if and only if $f \in L^p(\Omega; X)$ and there exist weak weak* partial derivatives $\partial_j f$ satisfying $\|\partial_j f(\cdot)\|_{(\text{Lip}_{z_0}(X))^*} \in L^p(\Omega), \quad j=1,\ldots,n$

Significance:

• Resolves the measurability problem of weak weak* derivative norms raised in 5, Subsection 3.4
• Provides an intrinsic Sobolev space characterization without requiring isometric embedding

2. Lipschitz Approximation (Lemma 3.10)

For $f \in W^{1,p}(\Omega; V)$ (where $V$ is a Banach space), for any $\varepsilon > 0$, there exists a Lipschitz mapping $F$ such that $|\{x: f(x) \neq F(x)\}| < \varepsilon, \quad \|f - F\|_{W^{1,p}} < \varepsilon$

3. Metric Differentiability in $W^{1,p}$ Topology (Theorem 4.4)

Mappings in $W^{1,p}(\Omega; X)$ are metric differentiable almost everywhere in the $W^{1,p}$ topological sense, i.e., $\left\|\frac{d(f(x+h\cdot), f(x))}{h} - \sigma_x(\cdot)\right\|_{W^{1,p}(B)} = o(1)$

Corollary 4.5: When $p > n$, functions in $W^{1,p}(\Omega; X)$ have continuous representatives and are metric differentiable almost everywhere

Special Results for Linear Target Spaces

Theorem 5.1 (General Banach Spaces)

For Lipschitz mappings $f: \Omega \to V$, there exists $\nabla^{**}f: \Omega \to B(\mathbb{R}^n; V^{**})$ such that for any separable subset $D^* \subset V^*$, almost everywhere $\lim_{y\to x} \frac{\langle v^*, f(y) - f(x) - \nabla^{**}f(x) \cdot (y-x) \rangle}{|y-x|} = 0, \quad \forall v^* \in D^*$

Theorem 5.2 (Dual Banach Spaces)

For Lipschitz mappings $f: \Omega \to V^*$, there exists $\nabla^*f: \Omega \to B(\mathbb{R}^n; V^*)$ with similar properties

Importance: These results show that even without complete weak* differentiability, similar differential properties can be obtained on separable subsets

Related Work

Historical Development

1. Kirchheim (1994): Establishes the metric Rademacher theorem, proving that Lipschitz mappings are metric differentiable almost everywhere
2. Ambrosio-Kirchheim (2000):
   • Simplifies the proof through isometric embedding into the dual of separable Banach spaces
   • Establishes the relationship between metric differentials and weak* differentials
3. Reshetnyak (1997, 1969):
   • Introduces metric-valued Sobolev spaces $W^{1,p}(\Omega; X)$
   • Studies generalized derivatives and almost everywhere differentiability
4. Hajłasz-Tyson (2008), Hajłasz (2009, 2011):
   • Studies Lipschitz density of Sobolev mappings between metric spaces
   • Points out the dependence of Sobolev mapping properties on isometric embeddings
5. Creutz-Evseev (2023, 2024):
   • First systematically introduces weak weak* derivatives in [5]4
   • Studies weak differentiability of absolutely continuous mappings
6. Gigli-Pasqualetto-Soultanis (2020):
   • Proposes an alternative linear interpretation of differentials for metric-valued Sobolev mappings
   • Theorem 4.7 of this paper is consistent with their Theorem 4.7

Innovations of This Paper

1. Conceptual Innovation: Systematically develops weak weak* derivative theory, providing an intrinsic method independent of isometric embeddings
2. Technical Breakthroughs:
   • Proves measurability of weak weak* derivative norms (resolves an open problem in 5)
   • Establishes precise norm representation of metric derivatives
3. Application Extensions:
   • Applies the theory to metric differentiability in $W^{1,p}$ topology
   • Studies the case of general Banach space targets

Conclusions and Discussion

Main Conclusions

1. Core Achievement: Through the weak weak* derivative framework, provides an intrinsic linear representation for metric differentials, given by $\text{md}(f,x)(\nu) = \rho(\nabla^{\circ}f(x) \cdot \nu)$ where $\rho$ is a seminorm on $(\text{Lip}_{z_0}(X))^*$
2. Theory Refinement:
   • When the target space is separable, $\rho$ is a norm
   • For Sobolev mappings, establishes a complete differential theory
   • Proves metric differentiability almost everywhere in the $W^{1,p}$ topology
3. Unified Framework: Unifies Kirchheim's metric differentiability theory with the Ambrosio-Kirchheim linearization method

Limitations

1. Seminorm Issue: When the target space is non-separable, $\rho$ is only a seminorm and may vanish on certain subspaces
2. Assumption Conditions:
   • Main results (Lemma 4.1) depend on Lipschitz continuity and metric differentiability of the mapping
   • Does not completely resolve the case under weaker assumptions
3. Computational Complexity: Computing $\rho$ in practice requires essential norming sequences, which may be difficult for non-separable spaces
4. Embedding Dependence: The proof of Theorem 4.4 still uses isometric embedding techniques (acknowledged by the author)

Future Directions

1. Open Problems: Can relationship (13) be established without assuming metric differentiability?
2. Generalization Directions:
   • Study the theory on more general metric measure spaces
   • Explore connections with differentiability of mappings on Carnot groups
3. Application Prospects:
   • Applications to rectifiable set theory in geometric measure theory
   • Applications in optimal transport theory

In-Depth Evaluation

Strengths

1. Theoretical Depth:
   • Resolves fundamental problems in differential theory of metric space-valued mappings
   • Provides an intrinsic method independent of specific embeddings
   • Rigorous proofs with clear logical structure
2. Technical Innovation:
   • Clever use of Banach limits to construct weak weak* derivatives
   • Introduction of essential norming sequences simplifies many proofs
   • Application of maximal function estimates in Lemma 3.9 demonstrates sophistication
3. Completeness:
   • Complete system from basic definitions to profound theorems
   • Covers multiple aspects of Sobolev space theory
   • Clear connections with existing literature
4. Writing Quality:
   • Clear structure, progressing from motivation to applications
   • Comprehensive background material, accessible to non-specialists
   • Sufficient proof details

Weaknesses

1. Abstraction:
   • Highly abstract functional analysis theory, lacking concrete examples
   • No numerical examples illustrating practical applications
2. Technical Dependence:
   • Some proofs (e.g., Theorem 4.4) still depend on isometric embeddings
   • Somewhat contradicts the stated goal of "intrinsic methods"
3. Computability:
   • Practical computational methods for theoretical results are unclear
   • Lacks detailed explanation of how to construct essential norming sequences for specific mappings
4. Application Scope:
   • Primarily focuses on establishing theoretical framework
   • Lacks concrete applications in geometric or analytical problems

Impact

1. Academic Contribution:
   • Provides new tools for the intersection of metric geometry and functional analysis
   • Resolves open problems posed in 5
   • Expected to be widely cited by researchers in metric space analysis theory
2. Theoretical Value:
   • Refines metric-valued Sobolev space theory
   • Lays foundation for further study of calculus on non-smooth spaces
   • Unifies multiple existing theoretical frameworks
3. Practical Value:
   • Reference value for researchers studying non-Euclidean geometry and sub-Riemannian geometry
   • Potential applications in metric space-valued functions in data analysis
4. Reproducibility:
   • Pure theoretical paper, all results verifiable through proofs
   • No numerical experiment reproducibility issues

Applicable Scenarios

1. Theoretical Research:
   • Differential geometry on metric spaces
   • Analysis of Banach space-valued functions
   • Geometric measure theory
2. Related Fields:
   • Optimal transport theory
   • Sub-Riemannian geometry with Sobolev spaces
   • Analysis on metric measure spaces
3. Potential Applications:
   • Shape analysis and computer vision with metric spaces
   • Manifold learning in data science
   • Non-smooth spacetime geometry in physics

Key References

1 L. Ambrosio, B. Kirchheim. Rectifiable sets in metric and Banach spaces. Math. Ann., 318(3):527–555, 2000.

14 B. Kirchheim. Rectifiable metric spaces: Local structure and regularity of the Hausdorff measure. Proc. Am. Math. Soc., 121(1):113–123, 1994.

16 Yu. G. Reshetnyak. Sobolev-type classes of functions with values in a metric space. Sib. Math. J., 38(3):657–675, 1997.

5 P. Creutz, N. Evseev. Weak differentiability of metric space valued Sobolev maps. Proc. Amer. Math. Soc, to appear.

6 N. Gigli, E. Pasqualetto, E. Soultanis. Differential of metric valued Sobolev maps. J. Funct. Anal., 278(6):24, 2020.

Overall Assessment: This is a high-quality theoretical mathematics paper making significant contributions to differential theory of metric space-valued mappings. By introducing the weak weak* derivative framework, the author successfully provides an intrinsic linear representation for metric differentials, avoiding dependence on specific isometric embeddings. The paper features rigorous proofs, complete structure, and is expected to have lasting impact on metric geometry and functional analysis. The main limitations are the abstraction of the theory and lack of concrete application examples, but these are acceptable for foundational theoretical work.