Viscosity Solution Frameworks

• Viscosity solution frameworks are non-variational methods that define generalized solutions for nonlinear PDEs using test function inequalities.
• They bypass classical differentiability by comparing upper and lower semicontinuous solutions, ensuring stability under weak convergence and uniqueness via comparison principles.
• Applications include control problems, stochastic games, and path-dependent or nonlocal PDEs, with extensions to strong-viscosity and measure-driven formulations.

A viscosity solution framework is a nonlinear, non-variational analytic approach to the existence, characterization, and stability of generalized solutions to broad classes of nonlinear PDEs, variational inequalities, control problems, stochastic games, and related models. This framework interprets PDEs—often fully nonlinear, degenerate, or singular—in terms of inequalities under smooth test functions touching the candidate solution from above or below. This approach bypasses classical differentiability and is robust under weak convergence, allowing for existence and uniqueness under minimal regularity. Viscosity solution theory encompasses classical PDEs, path-dependent PDEs, integro-PDEs, dynamic programming limits, and nonlocal/measure-driven equations. Several generalizations—such as strong-viscosity, mixed Aleksandrov–viscosity, or abstract DPP-based notions—further extend the method’s reach.

1. Foundational Definitions and Methodology

Consider a nonlinear PDE presented in abstract form: $$F(x,u(x),\nabla u(x),D^2 u(x)) = 0 \quad \text{in}~\Omega.$$ A function $u$ is an upper semicontinuous (USC) viscosity subsolution if, for every $C^2$ test function $\phi$ for which $u-\phi$ attains a local maximum at $x_0\in\Omega$, it holds that

$$F(x_0, u(x_0), \nabla\phi(x_0), D^2\phi(x_0)) \leq 0.$$

A lower semicontinuous (LSC) viscosity supersolution reverses the maximum/minimum and inequality. A viscosity solution is a function both a subsolution and a supersolution with suitable boundary behavior (Zhang et al., 2014, Eyssidieux et al., 2010, Cosso et al., 2015).

For path-dependent PDEs, the jet notion and test functionals are replaced by Dupire's vertical and horizontal functional derivatives on path space, as in (Peng, 2011, Cosso et al., 2015).

Key Properties

• Viscosity solutions do not require $u$ to be differentiable anywhere in $\Omega$.
• The framework is strictly comparison-based and maximally non-variational.
• It admits robust stability under uniform limits, as in Barles–Souganidis theory.

2. Classical and Strong-Viscosity Solutions

For semilinear or degenerate parabolic PDEs (e.g., Kolmogorov type), the viscosity framework coexists with classical and strong-viscosity notions.

Classical Solutions

A classical solution $u\in C^{1,2}$ solves the PDE pointwise, but existence is rarely guaranteed for fully nonlinear degenerate problems.

Standard Viscosity Solutions

Continuous $u$0 fulfilling the test-function inequality at local maxima/minima give the viscosity solution. Comparison principles (uniqueness) are established via doubling variables and Ishii–Jensen's lemma (Zhang et al., 2014, Eyssidieux et al., 2010).

Strong-Viscosity Solutions

A function $u$1 is a strong-viscosity solution if there exists a sequence of classical solutions $u$2 to perturbed, smooth-coefficient PDEs with $u$3 uniformly on compacts, and the coefficients converge pointwise. Every strong-viscosity solution is a viscosity solution, and uniqueness extends via stochastic representation (Feynman–Kac, BSDEs), e.g., $u$4 for the associated backward SDE (Cosso et al., 2015).

Solution Class	Regularity	Definition Approach	Uniqueness Mechanism
Classical	$u$5	Pointwise equality	PDE analytics
Viscosity (Crandall-Lions)	Continuous	Test function inequality	Comparison principle
Strong-viscosity	Uniform limit of smooth	Perturbation limit	BSDE/Feynman–Kac

3. Path-Dependent and Nonlocal Generalizations

Viscosity methods extend to fully nonlinear parabolic path-dependent PDEs (PPDEs) via horizontal and vertical functional derivatives, leading to robust comparison principles (e.g., Peng's left-frozen maximization for path spaces) (Peng, 2011). In these contexts, the solution is defined with respect to functionals on the path space $u$6, with the test-jet adapted for noncompact path spaces.

Nonlocal PDEs (e.g., with Lévy jumps) incorporate integro-differential terms within the viscosity formalism by evaluating the nonlocal operator on the test-function at the global maximum (cf. Imbert–Barles, and the IPDE approach to jump-diffusion games) (Biswas, 2010).

Mixed Aleksandrov–viscosity frameworks combine measure constraints (subgradient mass) with standard viscosity for fully nonlinear equations with singular right-hand sides, as in semi-discrete Monge–Ampère problems (Benamou et al., 2014).

4. Comparison Principles, Stability, and Regularity

The critical tool is the comparison (maximum) principle: if $u$7 is a USC subsolution and $u$8 an LSC supersolution, $u$9 in $C^2$0 when the PDE and boundary conditions are compatible. Proofs employ the doubling-of-variables, penalization, and local/jet estimates. This mechanism secures uniqueness, under suitable structural ellipticity, monotonicity, and regularity (Eyssidieux et al., 2010, Zhang et al., 2014, Cosso et al., 2015, Courte et al., 2020, Peng, 2011, Teso et al., 10 Feb 2026).

Stability follows: limits of sequences of viscosity sub/supersolutions (under uniform or half-relaxed convergence) are again viscosity sub/supersolutions of the limit equation, even under weak consistency (Barles–Souganidis framework), crucial in numerical and dynamic programming contexts (Teso et al., 10 Feb 2026).

Viscosity solutions possess optimal regularity given structural bounds: global or local Lipschitz regularity may be obtained for value functions and control problems under mild coercivity (Pablo et al., 2020). In degenerate complex Monge–Ampère equations, the viscosity solution coincides with the (pluripotential) measure solution and is continuous under weak integrability assumptions (Eyssidieux et al., 2010).

5. Applications to Control, Games, Optimization, and Geometric PDEs

The viscosity framework underpins solution theory for stochastic control (Hamilton–Jacobi–Bellman equations), stochastic differential games (Hamilton–Jacobi–Isaacs or IPDEs), regime-switching and optimal stopping variational inequalities, and quasi-variational inequalities for impulse/switching control (Zhang et al., 2014, Biswas, 2010, Cosso et al., 2015).

Notable intersections include:

• Characterization of optimal stopping, barriers, and game value functions via variational inequalities (min/max operators, interconnected systems) (Zhang et al., 2014, Hirsch et al., 11 Oct 2025).
• Regularity and structure of value functions in parametric and Banach-space optimization, with connections to generalized gradients (Clarke subdifferentials) (Pablo et al., 2020).
• Nonlinear geometric flows, including rate-independent mean curvature flows, vanishing viscosity methods, and level set equations, where the viscosity approach addresses both rate-independent and energetic solution branches (Courte et al., 2020).
• Degenerate/singular elliptic equations on manifolds, the complex Monge–Ampère equation, and pluripotential theory, with demonstrable uniqueness via the viscosity comparison and continuity via the $C^2$1 regularity (Eyssidieux et al., 2010).

6. Viscosity Solution Frameworks in Numerical Approximation and Dynamic Programming

Dynamic programming and game-theoretic schemes often generate only monotone, possibly discontinuous, approximation schemes ("DPPs"). Recent viscosity frameworks dispense with classical measurability, defining viscosity sub- and supersolutions with respect to monotone operators acting on test functions, and giving comparison between classical strict supersolutions and viscosity subsolutions (Teso et al., 10 Feb 2026).

In this context, convergence and existence are obtained by stability and weak consistency: any limit of sub/supersolutions of consistent DPPs is a viscosity solution of the corresponding PDE. Asymptotic expansions in the approximation parameter yield higher-order corrections to the limit equation (Teso et al., 10 Feb 2026).

Viscosity-based approaches have also been developed for computational methods, such as PINN-based neural approximations of HJB equations, where convexity constraints on the neural representation ensure convergence to the viscosity solution, avoiding spurious minimizers typical in nonconvex landscapes (Liu et al., 2023).

7. Extensions, Generalizations, and Limitations

Several major extensions exist:

• Path-dependent and stochastic control: functional derivatives, quasi-sure settings, and G-expectation martingale theory (Cosso et al., 2015, Peng, 2011).
• Measure-data equations: coupling Aleksandrov (mass constraint) and viscosity (test function) at Dirac masses, ensuring convergence even for singular measure right-hand sides (Benamou et al., 2014).
• Rate-independent evolution, doubly-nonlinear inclusions, and variational evolutions are handled by a set-valued generalization of the viscosity definition, where the dissipation law enters in the jet-based inequality (Courte et al., 2020).
• Control barrier functions and Hamilton–Jacobi–Reachability safety theory are unified under viscosity CBFs, extending invariance and barrier guarantees to non-smooth domains and value functions (Hirsch et al., 11 Oct 2025).
• Statistical convergence paradigms for singular limits, such as $C^2$2-convergence and Young measures in vanishing viscosity limits for fluid systems, with all observables recovered in the viscosity sense (Feireisl et al., 2021).

Limitations include the requirement of comparison principles (for uniqueness), possible loss of strong regularity, and challenge of constructing explicit sub- and supersolutions in highly irregular or nonlocal settings. Further, numerical methods must enforce monotonicity to guarantee consistency with viscosity solution theory.

---

References:

• "Strong-viscosity Solutions: Semilinear Parabolic PDEs and Path-dependent PDEs" (Cosso et al., 2015)
• "Optimal stopping problems with regime switching: a viscosity solution method" (Zhang et al., 2014)
• "Viscosity solutions to degenerate Complex Monge-Ampère equations" (Eyssidieux et al., 2010)
• "A viscosity solution approach to regularity properties of the optimal value function" (Pablo et al., 2020)
• "A viscosity framework for computing Pogorelov solutions of the Monge-Ampere equation" (Benamou et al., 2014)
• "Viscosity Solutions for Doubly-Nonlinear Evolution Equations" (Courte et al., 2020)
• "Note on Viscosity Solution of Path-Dependent PDE and G-Martingales" (Peng, 2011)
• "On zero-sum Stochastic Differential Games with Jump-Diffusion driven state: A viscosity solution framework" (Biswas, 2010)
• "PINN-based viscosity solution of HJB equation" (Liu et al., 2023)
• "Solution multiplicity and effects of data and eddy viscosity on Navier-Stokes solutions inferred by physics-informed neural networks" (Wang et al., 2023)
• "Approximating viscosity solutions of the Euler system" (Feireisl et al., 2021)
• "A Viscosity Framework for Dynamic Programming Principles and Applications" (Teso et al., 10 Feb 2026)
• "Viscosity CBFs: Bridging the Control Barrier Function and Hamilton-Jacobi Reachability Frameworks in Safe Control Theory" (Hirsch et al., 11 Oct 2025)