Non-convex coercive Hamilton-Jacobi equations: Guerand's relaxation revisited

Basic Information

• Paper ID: 2309.08224
• Title: Non-convex coercive Hamilton-Jacobi equations: Guerand's relaxation revisited
• Authors: Nicolas Forcadel, Cyril Imbert, Régis Monneau
• Classification: math.AP (Mathematical Analysis)
• Publication Date: September 18, 2023
• Paper Link: https://arxiv.org/abs/2309.08224

Abstract

This paper investigates evolutionary Hamilton-Jacobi equations defined on a domain with supplementary boundary conditions. The Hamiltonian function is coercive but neither convex nor quasiconvex. The authors analyze boundary conditions in the viscosity solution sense, based on the study of evolutionary boundary conditions. More precisely, the authors provide a new formulation of the relaxed boundary conditions derived by J. Guerand, offering a new perspective that reveals the connection between the relaxation operator and the classical Godunov flux from conservation law theory. The authors apply the method to two classical boundary value problems, demonstrating that relaxed Neumann boundary conditions can be expressed using the Godunov flux, while relaxed Dirichlet boundary conditions reduce to a boundary obstacle problem related to the lower non-increasing envelope of the Hamiltonian function.

Research Background and Motivation

Importance of the Problem

Hamilton-Jacobi equations have broad applications in optimal control, differential games, image processing, and other fields. When solving equations on bounded domains, the treatment of boundary conditions becomes critical. In particular, when characteristics reach the boundary, boundary conditions may conflict with the equation itself.

Limitations of Existing Methods

1. Restriction to convex Hamiltonians: Early research focused primarily on convex Hamiltonian functions, limiting the scope of the theory
2. Compatibility of boundary conditions: Traditional methods struggle with cases where boundary conditions are incompatible with the equation
3. Complexity of non-convex cases: The structure of relaxed boundary conditions for non-convex Hamiltonians is more intricate, lacking a unified theoretical framework

Research Motivation

This paper aims to provide a unified boundary condition relaxation theory for non-convex coercive Hamiltonian functions, establish deep connections with conservation law theory, and provide explicit expressions for specific boundary value problems.

Core Contributions

1. New relaxation operator formula: Proposes a new expression for the Guerand relaxation operator, simplifying theoretical analysis
2. Connection with Godunov flux: Establishes a deep link between the relaxation operator and the classical Godunov flux
3. Unified theoretical framework: Proves the equivalence of three different relaxation operators (Guerand operator, new operator, Godunov operator)
4. Solution of specific boundary value problems: Provides explicit expressions for relaxed boundary conditions in Neumann and Dirichlet problems
5. Multi-dimensional generalization: Extends one-dimensional results to multi-dimensional cases

Detailed Methodology

Problem Formulation

Consider the following Hamilton-Jacobi equation:

{
  u_t + H(t,x,Du) = 0,  t > 0, x ∈ Ω
  boundary conditions,    t > 0, x ∈ ∂Ω
}

where Ω is a C¹ domain in ℝ^d and H is a coercive but non-convex Hamiltonian function.

Core Theoretical Construction

1. Relaxation Operator Definition

For the one-dimensional case, define upper and lower relaxation operators:

RF₀(p) := sup_{q≥p} (F₀ ∧ H)(q)
RF₀(p) := inf_{q≤p} (F₀ ∨ H)(q)

Then define the relaxation operator:

RF₀(p) = {
  RF₀(p)  if F₀(p) ≥ H(p)
  RF₀(p)  if F₀(p) ≤ H(p)
}

2. Multi-dimensional Generalization

In the multi-dimensional case, achieved by freezing tangential variables:

RF₀(t,x,p) = sup_{ρ≥0} (F₀ ∧ H)(t,x,p-ρn)
RF₀(t,x,p) = inf_{ρ≤0} (F₀ ∨ H)(t,x,p-ρn)

3. Characteristic Point Theory

Define positive and negative characteristic points:

• p is a positive characteristic point of F₀: H(p) = F₀(p) and H > H(p) on (p, p+ε)
• p is a negative characteristic point of F₀: H(p) = F₀(p) and H < H(p) on (p-ε, p)

Technical Innovations

1. Simplified relaxation formula: The new formula is derived directly from the weak viscosity solution definition, avoiding complex limiting point constructions
2. Characteristic point analysis: Simplifies the test function set through characteristic point theory
3. Operator equivalence proof: Rigorously proves the equivalence of three relaxation operators
4. Conservation law connection: Establishes the link between boundary conditions for Hamilton-Jacobi equations and those for scalar conservation laws

Main Theorems

Theorem 1.1 (Relaxed Boundary Conditions)

Assume H, F₀ are continuous, H is coercive, F₀ is semi-coercive and monotone non-decreasing with respect to p·n. Then there exists a continuous semi-coercive function RF₀ such that u is a weak viscosity solution of (1.1) if and only if it is a strong viscosity solution of:

{
  u_t + H(t,x,Du) = 0,     t > 0, x ∈ Ω
  u_t + RF₀(t,x,Du) = 0,   t > 0, x ∈ ∂Ω
}

Theorem 1.3 (Operator Equivalence)

Under appropriate assumptions, RF₀ = JF₀, i.e., the new relaxation operator equals the Guerand operator.

Theorem 1.5 (Neumann Problem)

The relaxed boundary condition for the Neumann problem is given by the Godunov flux:

N(t,x,p) = {
  max{H(t,x,p-ρn) : ρ ∈ [0,p·n(x)+h(t,x)]}  if p·n(x)+h(t,x) ≥ 0
  min{H(t,x,p-ρn) : ρ ∈ [p·n(x)+h(t,x),0]}  if p·n(x)+h(t,x) ≤ 0
}

Theorem 1.6 (Dirichlet Problem)

The relaxed boundary condition for the Dirichlet problem is a boundary obstacle problem:

max{u-g, u_t + H⁻(t,x,Du)} = 0

where H⁻(t,x,p) = inf_{ρ≤0} H(t,x,p-ρn(x)) is the lower non-increasing envelope of the Hamiltonian function.

Experimental Setup and Verification

Theoretical Verification

The paper primarily verifies theoretical results through rigorous mathematical proofs:

1. Operator property verification: Proves continuity, monotonicity, and semi-coercivity of the relaxation operator
2. Equivalence proofs: Proves equivalence of different operator expressions through characteristic point analysis
3. Boundary condition analysis: Verifies correctness of relaxed boundary conditions using test function methods

Concrete Examples

Example 1.2 (Completely degenerate case): When F₀ is a constant A, the relaxation function is:

RF₀ = max(A, H⁻)

Counterexample 3.16: Constructs a counterexample showing that conclusions may fail without semi-coercivity or weak continuity conditions.

Related Work

Historical Development

1. Viscosity solution theory: The viscosity solution theory established by Crandall-Evans-Lions provides the foundation for boundary condition analysis
2. Convex case research: Imbert-Monneau and others studied network problems with convex Hamiltonians
3. Non-convex generalization: Guerand pioneered the study of relaxation theory for one-dimensional non-convex cases

Uniqueness of This Paper's Contribution

1. Multi-dimensional generalization: Extends Guerand's one-dimensional theory to multiple dimensions
2. New theoretical perspective: Establishes connections with conservation law theory
3. Unified framework: Provides a unified method for handling different boundary conditions

Conclusions and Discussion

Main Conclusions

1. Theoretical unification: Proves equivalence of three relaxation operator expressions, providing a unified theoretical framework
2. Computational simplification: The new formula is more suitable for numerical computation and theoretical analysis
3. Application extension: Provides theoretical foundation for numerical methods for non-convex Hamilton-Jacobi equations

Limitations

1. Coercivity requirement: The theory requires coercivity assumptions on the Hamiltonian function
2. Regularity conditions: The C¹ regularity requirement on the boundary may be too strong for some applications
3. Weak continuity: Additional weak continuity conditions are needed when semi-coercivity is absent

Future Directions

1. Numerical methods: Develop efficient numerical algorithms based on the new theory
2. Application extension: Apply the theory to more practical problems
3. Relaxation of regularity: Study the theory under weaker regularity conditions

In-Depth Evaluation

Strengths

1. Theoretical depth: Provides profound mathematical insights and establishes connections between different theories
2. Methodological innovation: The new relaxation operator formula is more intuitive and practical
3. Completeness: The theoretical framework is complete, covering major boundary condition types
4. Rigor: Mathematical proofs are rigorous with clear logic

Weaknesses

1. Lack of numerical verification: Lacks numerical experiments to verify the practical effectiveness of the theory
2. Computational complexity: Does not analyze the computational complexity of the new method
3. Practical applications: Lacks case studies in specific application domains

Impact

1. Theoretical contribution: Makes important contributions to the theory of non-convex Hamilton-Jacobi equations
2. Methodological significance: The unified framework may influence research methods in related fields
3. Application potential: Provides theoretical foundation for numerical methods and practical applications

Applicable Scenarios

1. Optimal control: Optimal control problems with non-convex cost functions
2. Differential games: Value function computation in multi-player games
3. Numerical analysis: Design of numerical methods for Hamilton-Jacobi equations
4. Image processing: Image processing problems based on variational methods

References

The paper cites important literature in the field, including:

• Guerand's pioneering work 15,16
• Imbert-Monneau's network theory 17,18
• Lions-Souganidis's non-convex theory 25,26
• Classical viscosity solution theory 12,13,14

Summary: This is a high-quality theoretical mathematics paper that makes important contributions to the boundary condition theory for non-convex Hamilton-Jacobi equations. The paper's main value lies in providing a unified theoretical framework and more practical operator expressions. Although it lacks numerical verification, it establishes a solid theoretical foundation for subsequent applied research.