Metric viscosity solutions and distance-like functions on the Wasserstein space

Basic Information

• Paper ID: 2311.10618
• Title: Metric viscosity solutions and distance-like functions on the Wasserstein space
• Authors: Huajian Jiang, Xiaojun Cui
• Classification: math.AP (Analysis of PDEs)
• Publication Date: January 3, 2025 (arXiv v4)
• Paper Link: https://arxiv.org/abs/2311.10618

Abstract

This paper investigates metric viscosity solutions, distance-like functions, and their interrelationships on the Wasserstein space. On non-compact complete Riemannian manifolds, viscosity solutions of the eikonal equation |∇u|_g = 1 are precisely distance-like functions, which is crucial for understanding geometric and topological properties. The authors generalize this theory to general metric spaces, particularly to the Wasserstein space P_p(X) where X is a complete, separable, locally compact, and non-compact geodesic space, and provide two methods for constructing strong metric viscosity solutions.

Research Background and Motivation

1. Problem Background

In classical differential geometry, there exists a profound connection between viscosity solutions of the eikonal equation |∇u|_g = 1 and distance-like functions. Distance-like functions represent distances to infinity, and their 1-Lipschitz property makes them viscosity subsolutions of the eikonal equation, while the existence of negative gradient curves makes them viscosity supersolutions.

2. Research Challenges

Classical viscosity solution theory heavily relies on differential structure and cannot be directly extended to general metric spaces. Although various definitions of metric viscosity solutions exist, their mutual relationships and connections with distance-like functions remain unclear.

3. Importance of Wasserstein Space

The Wasserstein space occupies a fundamental position in optimal transport theory with widespread applications in geometric inference, generative adversarial networks, clustering analysis, mean field games, and other fields. Despite the lack of local compactness in Wasserstein space, its rich geometric structure makes the study of metric viscosity solutions on it highly significant.

4. Research Motivation

This paper aims to:

• Establish equivalence between metric viscosity solutions and distance-like functions on metric spaces
• Construct strong metric viscosity solutions on Wasserstein space
• Investigate geometric properties and existence of these solutions

Core Contributions

1. Established equivalence between metric viscosity solutions and distance-like functions: On complete unbounded length spaces, metric viscosity solutions are precisely dl_G-functions (Theorem 1.5)
2. Provided existence of ε-negative gradient curves: Proved that any metric viscosity solution admits global ε-negative gradient curves (Theorem 1.7)
3. Constructed strong metric viscosity solutions on Wasserstein space:
   • Constructed dl_C-functions via sequences of closed sets satisfying the (CS) condition (Theorem 1.12)
   • Constructed solutions on Wasserstein space from metric viscosity solutions in the ambient space (Theorem 1.13)
4. Introduced the (CS) condition: A key technical condition for constructing strong metric viscosity solutions on Wasserstein space
5. Provided counterexamples and stability analysis: Demonstrated that dl_C-functions are not always metric viscosity solutions and analyzed solution stability

Methodology Details

Problem Formulation

Study metric viscosity solutions of the eikonal equation |∇u|(y) = 1 on metric spaces, where |∇u|(y) := lim sup_{x→y} |u(x)-u(y)|/d(x,y) is the generalized gradient norm.

Core Definitions

Metric Viscosity Solution (Definition 1.1): On an open set Ω of a length space (Y,d), a locally Lipschitz function u is called a metric viscosity solution if for all y∈Ω, |∂u|(y) = 1, where $|∂u|(y) := \limsup_{x∈Ω,x→y} \frac{(u(y)-u(x))^+}{d(y,x)}$

Distance-like Functions:

• dl_C-function (Definition 1.2): For a sequence of closed sets {H_n}, the function u(y) = lim_{n→∞}d(y,H_n) - c_n
• dl_G-function (Definition 1.3): A function satisfying u(y) = c + d(y, u^{-1}(-∞,c]) for all c < u(y)

Key Technical Tools

1. Partial Order Relation (Definition 3.3): For a 1-Lipschitz function u and δ > 0, define x ≼_δ y if and only if u(x) - u(y) ≥ δd(x,y)

2. (CS) Condition (Definition 1.9): A sequence {y_n} satisfies the (CS) condition if there exist σ > 0 and unit-speed geodesics μ_n such that ∂B_σ(y_0) ∩ _{n∈ℕ} has a convergent subsequence

Proof Strategy of Main Theorems

Proof of Theorem 1.5:

• (i)⇒(ii): Using Lemma 3.5, prove the dl_G property through construction of monotone decreasing chains
• (ii)⇒(i): Verify the sphere minimization condition directly using the dl_G property
• Key step: Use Zorn's lemma in Lemma 3.4 to construct minimal elements

Proof of Theorem 1.12:

• Utilize the (CS) condition to ensure existence of co-rays (Lemma 4.4)
• Obtain convergence of geodesic sequences via Ascoli-Arzelà theorem
• Prove that the limit curve is a negative gradient ray

Experimental Setup

Theoretical Verification

This paper is primarily theoretical research, with verification through concrete examples:

Example 2: Constructed a counterexample where dl_C-function is not a metric viscosity solution

• Set ω_n = (1-1/n²)δ_0 + (1/n²)δ_{n²} ∈ P_2(ℝ)
• The function u_n(ω) = W_2(ω,ω_n) - W_2(δ_0,ω_n) is a metric viscosity solution
• But the limit function u ≡ 0 is not a metric viscosity solution

Examples 3-5: Verification of (CS) condition

• Example 3: Sequences on rays satisfy the (CS) condition
• Example 4: Sequences of Dirac measures satisfy the (CS) condition
• Example 5: Specially constructed sequences do not satisfy the (CS) condition

Constructive Proofs

Strong metric viscosity solutions are constructed in two ways:

1. Starting from sequences of closed sets satisfying the (CS) condition
2. Starting from metric viscosity solutions on the ambient space X

Experimental Results

Main Theoretical Results

Equivalence Theorem (Theorem 1.5): On complete unbounded length spaces, the following are equivalent:

• u is a metric viscosity solution
• u is a dl_G-function
• u satisfies the sphere minimization condition: u(y) = inf_{x∈∂B_r(y)}{u(x) + d(x,y)}

Existence Theorem (Theorem 1.7): Any metric viscosity solution admits global ε-negative gradient curves that tend to infinity

Construction Theorems (Theorems 1.12-1.13):

• dl_C-functions satisfying the (CS) condition are strong metric viscosity solutions
• Metric viscosity solutions on the ambient space induce strong metric viscosity solutions on Wasserstein space

Important Findings

1. Failure of Stability: Metric viscosity solutions are unstable under pointwise convergence (Example 2), but stable under local uniform convergence (Proposition 3.14)
2. Representation Formula (Proposition 4.8): Strong metric viscosity solutions can be represented using Busemann functions: $u(ω) = \inf_{γ∈N(u)}[u(γ(0)) + b_γ(ω)]$
3. Weak KAM Property (Proposition 4.6): On non-branching spaces, negative gradient rays are unique

Related Work

Classical Theory

• Study of Busemann functions and horo-functions 4,5
• Viscosity solution theory on Riemannian manifolds 6,7,37
• Classical theory of eikonal equations 13

Extensions to Metric Spaces

• Various definitions of metric viscosity solutions 2,14,15,29,30,33
• Study of curve-based solutions 14
• Theory of Hamilton-Jacobi equations on metric spaces 16

Wasserstein Space Geometry

• Foundations of optimal transport theory 8
• Geometric properties of Wasserstein space 1,9
• Characterization of geodesics and rays 4

Conclusions and Discussion

Main Conclusions

1. Unified metric viscosity solutions with distance-like functions: Established complete equivalence on complete length spaces
2. Resolved existence problems on Wasserstein space: Provided two methods for constructing strong metric viscosity solutions
3. Revealed geometric structure: Characterized geometric properties of Wasserstein space through the (CS) condition

Limitations

1. Verification of (CS) condition: Verifying the (CS) condition in practical applications may be difficult
2. Absence of local compactness: The non-local compactness of Wasserstein space causes certain classical results to fail
3. Stability issues: Failure of stability under pointwise convergence limits the reliability of numerical computation

Future Directions

1. Numerical Methods: Develop effective algorithms for computing metric viscosity solutions
2. Applied Research: Explore applications in machine learning and optimal transport
3. Theory Generalization: Extend to more general metric measure spaces

In-Depth Evaluation

Strengths

1. Theoretical Depth: Established a complete framework for eikonal equation theory on metric spaces
2. Technical Innovation: The introduction of the (CS) condition is an important technical contribution for handling non-locally compact spaces
3. Constructive Results: Two different construction methods enhance the practical utility of the theory
4. Rigor: Detailed and rigorous proofs with appropriate counterexamples and clear logic

Weaknesses

1. Insufficient Application Orientation: Primarily pure theoretical research lacking concrete application scenarios
2. Computational Complexity: Insufficient analysis of computational complexity of construction methods
3. Limited Examples: Relatively few positive examples despite appropriate counterexamples

Impact

1. Theoretical Contribution: Establishes foundations for Hamilton-Jacobi theory on metric spaces
2. Cross-disciplinary Value: Connects optimal transport, differential equations, and metric geometry
3. Subsequent Research: Provides important theoretical tools for related fields

Applicable Scenarios

1. Optimal Transport Theory: Provides theoretical foundations for variational problems on Wasserstein space
2. Geometric Analysis: Study of asymptotic geometric properties of metric spaces
3. Machine Learning: Provides theoretical support for optimization problems on probability measure spaces

References

The paper cites 37 important references, covering:

• Classical textbooks on optimal transport theory 8,11
• Gradient flow theory on metric spaces 1
• Related research on eikonal equations 14,15,16
• Specialized studies on Wasserstein space geometry 4,5

Summary: This paper is an important contribution to eikonal equation theory on metric spaces, particularly establishing a complete theory of metric viscosity solutions on Wasserstein space. Although primarily theoretical research, it provides a solid mathematical foundation for related application fields.