Viscosity Solution Theory

• Viscosity solution theory is a framework that defines weak solutions for highly nonlinear and degenerate PDEs by replacing classical differentiability with comparison principles.
• It rigorously addresses complex problems in optimal transport, free boundary, and stochastic control, extending its impact to mathematical finance and geophysical flows.
• The theory underpins robust numerical methods, including monotone finite-difference schemes and vanishing-viscosity regularization, ensuring reliable convergence.

Viscosity solution theory provides a generalized, nonvariational framework for defining and analyzing weak solutions to highly nonlinear partial differential equations (PDEs), particularly those lacking classical differentiability or maximum principles. The viscosity approach is essential for fully nonlinear, degenerate elliptic or parabolic equations, free boundary and obstacle problems, and systems with constraints not directly resolvable by standard weak formulations. Its reach extends across deterministic and stochastic control, mass transport, mathematical finance, complex and real analysis, nonlinear PDE numerics, and geophysical flows.

1. Core Principles and Definitions

At its foundation, viscosity solution theory replaces the need for classical differentiability of the solution with comparison principles: a function is a subsolution (supersolution) if its difference with any smooth test function at a (local) maximum (minimum) satisfies the PDE’s inequality in an appropriate sense. This principle allows the meaningful study of degenerate or nonsmooth phenomena, including solutions with singularities, anisotropy, and nonlocal effects.

For a general second-order degenerate elliptic PDE

$$F(x, u(x), Du(x), D^2 u(x)) = 0 \quad \text{in } \Omega,$$

a bounded, upper semicontinuous $u$ is a viscosity subsolution if, for every $\varphi \in C^2$, at every local maximum $x_0$ of $u - \varphi$, the inequality

$$F^*(x_0, \varphi(x_0), D\varphi(x_0), D^2\varphi(x_0)) \le 0$$

holds (where $F^*$ is the upper semicontinuous envelope of $F$). The dual inequality defines supersolutions using the lower semicontinuous envelope. Viscosity solutions are functions which are both sub- and supersolutions.

For equations with constraints—such as convexity, local Hamilton–Jacobi conditions at boundaries, or inequalities arising from boundary value or obstacle problems—the operator $F$ can be discontinuous, and multiple “faces” of the PDE are encoded in the viscosity framework (Benamou et al., 2012).

2. Existence, Uniqueness, and Comparison Principles

The viscosity solution framework yields powerful comparison and uniqueness results under suitable monotonicity, ellipticity (degenerate or uniform), and boundary regularity conditions.

• Comparison Principle: If $u$ is a subsolution, $v$ is a supersolution, and $u \le v$ on the parabolic (or Dirichlet) boundary, then $u \le v$ in the domain. For instance, with convex target sets, degenerate elliptic Monge–Ampère operators, and oblique boundary data, existence and uniqueness of viscosity solutions (with convexity constraints) follow directly from comparison principles and Perron's method (Benamou et al., 2012).
• Existence: Viscosity solutions often exist via Perron's construction: suprema of subsolutions (or infima of supersolutions) are again viscosity solutions under stability and compactness assumptions (Courte et al., 2020). For equations admitting monotone/degenerate elliptic structure, a vanishing-viscosity regularization or monotone approximation yields convergent sequences to the unique viscosity solution (Almeida et al., 2020, Benamou et al., 2012).
• Regularity Theory: Viscosity solutions of uniformly elliptic (real or complex) equations possess Hölder regularity up to the boundary, with regularity exponents depending only on the ellipticity constants and integrability of the inhomogeneity (Sun, 2023).

3. Applications and Advanced Models

Viscosity solution theory has been adapted to a diverse array of advanced settings:

Domain	Key PDE/Problem Class	Unique Features
Optimal Transport	Monge–Ampère with global state constraints	Viscosity structure encodes convexity, boundary mapping
Nonlinear Option Pricing	Black–Scholes with Γ-dependent diffusion	Vanishing-viscosity ensures stability, existence
Stochastic Control and Games	HJB, Isaacs, and obstacle equations	Handles jump-diffusions, nonlocal terms, FBSDE coupling
Complex Geometry	Complex Monge–Ampère / Hessian equations	Complex Hermitian structure, determinant domination
Free Boundary/Stefan Problems	Two-phase parabolic and enthalpy formulations	Comparison via classical barriers, mushy region analysis
Rate-Independent/Bistable Dynamics	Doubly-nonlinear inclusions	Set-valued dissipation, hysteresis, discontinuities
Mean-Field / Wasserstein HJB	Infinite-dimensional, measure space	Stochastic, nonlocal, noncompact solution space
Numerical Analysis	Finite-difference, FV, PINN methods	Discrete monotonicity, filtered schemes, neural convexity

Specific examples:

• Monge–Ampère equation for optimal transport: Convex potential $u$ solves

$$\det(D^2 u) = \frac{\rho_X(x)}{\rho_Y(\nabla u(x))},\quad x \in X;\quad \nabla u(X)=Y,$$

with viscosity sub/supersolution definitions incorporating maps into the convex target, with boundary conditions formulated as Hamilton–Jacobi equations on the boundary (Benamou et al., 2012). Numerical convergence is established via monotone wide-stencil schemes and Barles–Souganidis theory.

• Complex Hessian equations on Hermitian manifolds: Solutions $u$ to

$F(\chi + dd^c u) = e^{G(z,u)},$

are viscosity sub(super)solutions if test perturbations from above (below) at each point satisfy spectral positivity in a cone and the associated PDE inequality; determinant domination provides the critical a priori estimate (Cheng et al., 2024, Sun, 2023).

• Doubly-nonlinear evolution equations: Viscosity solutions encompass inclusions or rate-independent flow of the form

$$F(x,t,u,u_t,\nabla u, D^2 u) \in \mathcal S(u_t) G(x,t,u,\nabla u),$$

using sub(super)jets and set-valued selection, handling discontinuity, hypoellipticity, and nonvariational systems (Courte et al., 2020).

• Stochastic HJB in Wasserstein spaces: Viscosity solutions extend to functional PDEs on spaces of probability measures, incorporating Lions’ derivatives and semimartingale decompositions, even in non-Markovian or mean-field interacting particle systems (Cheung et al., 2023).

4. Structure-Preserving Numerical Schemes

Viscosity solution theory directly informs the design and analysis of convergent numerical methods for fully nonlinear PDEs:

• Monotone/filtered finite-difference schemes: Consistency, monotonicity (degenerate ellipticity), and stability are the Barles–Souganidis pillars for ensuring discrete solutions converge uniformly to the unique viscosity solution as discretization parameters vanish (Benamou et al., 2012).
• Vanishing-viscosity regularization: Adding higher-order regularization to degenerate or non-uniformly parabolic problems produces families of classical solutions whose limits, as the regularization vanishes, are viscosity solutions (Almeida et al., 2020).
• Physics-informed neural networks for viscosity solutions: Optimization of neural networks under convexity or monotonicity constraints (via ICNNs or Hessian penalties) guarantees convergence to unique viscosity solutions of HJB equations, with numerical validation in high-dimensional settings (Liu et al., 2023).
• Ensemble/statistical numerics for measure-valued solutions: For hyperbolic and kinetic systems, the S-convergence framework allows computation of measure-valued (viscosity) solutions as statistical limits of finite-volume or finite-difference schemes (Feireisl et al., 2021).

5. Advanced Theoretical Developments and Regularity

• Regularity and structure: Viscosity solutions can exhibit regions of $C^1$ regularity (strips), determined by the uniqueness of maximizers in associated Hopf–Lax or Hopf-type representations, linked to semiconcavity/semiconvexity of the Hamiltonian and initial data (Hoang, 2016). Propagation and formation of singularities are analyzed through geometric and measure-theoretic techniques.
• Free boundary and multiphase flows: Comparison principles for viscosity solutions allow the rigorous handling of moving interfaces, mushy layers, and obstacles in multiphase Stefan and related problems, fully characterizing the equivalence with weak/variational (enthalpy) solutions when appropriate (Kim et al., 2010).
• Nonsmooth and discontinuous systems: The viscosity framework admits discontinuous solutions, accommodates set-valued inclusions, and captures companions of classical minimization/movement schemes (e.g., hysteresis and rate-independence), making it applicable to broad nonvariational and degenerate dynamics (Courte et al., 2020).

6. Connections, Extensions, and Research Directions

Viscosity solution theory continues to expand:

• Links to convex and nonsmooth analysis: The relation to generalized gradients (Clarke subdifferentials), variational inequalities, and verification/optimality conditions bridges viscosity PDEs with convex and set-valued analysis (Pablo et al., 2020).
• Extensions to stochastic, infinite-dimensional, and nonlocal contexts: New viscosity notions for infinite-dimensional Banach, Wasserstein, and path space settings have been developed for stochastic control, measure-valued dynamics, and games with nonlocal or jump processes (Biswas, 2010, Cheung et al., 2023).
• Unified frameworks: Current trends include multi-domain viscosity approaches (e.g. combined local/nonlocal operators, systems with nonstandard boundary/interface conditions, hybrid deterministic-stochastic systems), and seamless integration with energy/entropy-based solution concepts for evolutionary PDEs.

Viscosity solution theory thus constitutes a universal weak-solution framework for nonlinear PDEs, reconciling analytic, numerical, and variational aspects, and underpins modern approaches to a broad array of PDEs encountered in contemporary mathematics and its applications.