In classical calculus, a function must be differentiable to satisfy a partial differential equation (PDE). However, many real-world phenomena naturally produce solutions with "kinks," sharp corners, or sudden discontinuities. For instance, when tracking the reachable set of a dynamic system over time, the evolving boundary can develop sharp corners where classical derivatives cease to exist (Stanislav Mazurenko in Viscosity solutions to evolution problems of star-shaped reachable sets). Similarly, in image processing, sharp edges and contrast boundaries are fundamental features of the data, making standard derivative-based smoothing equations mathematically ill-defined at the most critical points of the image (Adina Ciomaga et al. in A Proof of Equivalence between Level Lines Shortening and Curvature Motion in Image Processing).
Even if an evolution equation begins with perfectly smooth initial conditions, non-linear interactions or degenerate noise can cause the solution to lose its regularity over time. This "roughening effect" creates non-smooth states that break traditional numerical methods (Martin Hairer et al. in Loss of regularity for Kolmogorov equations). To analyze and solve these equations rigorously, we need a framework that allows continuous, non-differentiable functions to serve as valid solutions. This is the purpose of the viscosity solution framework.
The Intuition: Testing from Above and Below
How do you evaluate a derivative at a point where the derivative does not exist? The core insight of viscosity solution theory is to shift the burden of differentiation away from the rough target function and onto a family of smooth test functions.
Imagine a non-smooth function $u(x)$ with a sharp, V-shaped kink pointing downward. We cannot compute the gradient $\nabla u(x)$ at the tip of the V. Instead of looking at $u(x)$ directly, we look at smooth, differentiable functions $\phi(x)$ that "touch" $u(x)$ at that point.
- Subsolutions (Testing from Below): If a smooth function $\phi$ touches $u$ from below at a point $x_0$ (meaning $\phi \le u$ locally and $\phi(x_0) = u(x_0)$), it acts as a local lower bound. We require that the derivatives of $\phi$ satisfy the PDE as an inequality (e.g., $F(x_0, u(x_0), \nabla \phi(x_0)) \le 0$).
- Supersolutions (Testing from Above): If a smooth function $\phi$ touches $u$ from above at a point $x_0$ (meaning $\phi \ge u$ locally and $\phi(x_0) = u(x_0)$), it acts as a local upper bound. We require that $\phi$ satisfies the PDE as the opposite inequality ($F(x_0, u(x_0), \nabla \phi(x_0)) \ge 0$).
A continuous function is a viscosity solution if it is both a subsolution and a supersolution. By sandwiching the non-smooth function between these smooth test functions, we can determine whether it satisfies the differential equation without ever differentiating the function itself.
As Antonin Chambolle et al. demonstrate in A Hölder infinity Laplacian, this method provides a rigorous way to handle highly non-linear and non-local operators involving suprema and infima. In more technical terms, this intuition can be formalized using mathematical constructs like parabolic superjets and subjets to establish weak solutions for non-linear equations that lack classical smoothness (Christian Beck et al. in On nonlinear Feynman–Kac formulas for viscosity solutions of semilinear parabolic partial differential equations).
Why It Works: Uniqueness and Stability
A major challenge with generalized or "weak" solutions in mathematics is that relaxing the definition of a derivative can introduce an infinite number of spurious, physically meaningless solutions. Viscosity solutions avoid this trap because they inherently preserve the maximum principle (often called the comparison principle).
The comparison principle states that if one valid subsolution starts below a supersolution on the boundary of a domain, it must remain below it throughout the entire interior. Fen Gu and Feida Jiang leveraged this property in Positive viscosity solutions of a third degree homogeneous parabolic infinity Laplace equation to prove the existence and uniqueness of solutions for highly degenerate non-linear equations where traditional calculus fails completely.
Because the framework ensures that solutions are stable and unique, researchers can construct global solutions using Perron's method. This approach involves defining a large set of admissible subsolutions and taking their upper envelope to find the exact viscosity solution (Yeoneung Kim et al. in State-Constraint Static Hamilton-Jacobi Equations in Nested Domains). This structural stability is what makes numerical approximation schemes provably convergent and resilient to noise in practical applications, such as reconstructing 3D geometry from digital photographs (Emmanuel Prados and Olivier D. Faugeras in Unifying Approaches and Removing Unrealistic Assumptions in Shape from Shading: Mathematics Can Help).
Applications Across Domains
The viscosity solution framework has become an essential tool across fields where non-smoothness is an intrinsic feature of the problem framing.
Stochastic Optimal Control and Games
In control theory, the value function (the minimum cost-to-go) often contains non-differentiable points due to strict state constraints or discontinuous costs. The authors of Weak Dynamic Programming for Generalized State Constraints and Stochastic Target Games and Dynamic Programming via Regularized Viscosity Solutions utilize probabilistic relaxation to derive Hamilton-Jacobi-Bellman (HJB) equations in the viscosity sense, bypassing long-standing technical hurdles like measurable selection.
This allows for the exact characterization of optimal control laws even for sparse, "bang-off-bang" control strategies where the controller must abruptly switch between fully on and completely off (Kaito Ito et al. in Sparse optimal stochastic control), or in adversarial game-theoretic models where an algorithm must defend against real-time malicious manipulation (Erhan Bayraktar et al. in Malicious Experts versus the multiplicative weights algorithm in online prediction).
Mathematical Finance
When modeling derivative pricing under uncertain volatility, the classic Black-Scholes model breaks down because it assumes volatility is a known constant. The Black–Scholes–Barenblatt (BSB) equation steps in to price derivatives when volatility is only known to stay within a certain range. This results in highly non-linear parabolic PDEs that lack smooth solutions but can be solved perfectly using continuous functions within the viscosity framework (Xinpeng Li and Weicheng Xu in On properties of solutions to Black–Scholes–Barenblatt equations).
High-Dimensional Systems and Advanced Physics
Viscosity solutions are also crucial for breaking the "curse of dimensionality" in numerical analysis. By proving that high-dimensional semilinear elliptic PDEs can be approximated without exponential complexity growth, the framework extends the practical utility of numerical models to massive financial and physical systems (Christian Beck et al. in Overcoming the curse of dimensionality in the numerical approximation of high-dimensional semilinear elliptic partial differential equations).
[Non-Smooth Real-World Problem]
│ (Classical derivatives fail due to kinks/shocks)
▼
[Apply Smooth Test Functions] ───► Test from Above (Supersolution)
│ ───► Test from Below (Subsolution)
▼
[Viscosity Solution Framework]
│ (Preserves Maximum/Comparison Principle)
▼
[Unique, Stable, and Convergent Numerical Solution]
This mathematical machinery allows engineers to optimize power control in highly dynamic 6G wireless channels driven by heavy-tailed, discontinuous Lévy noise (Mengqi Li et al. in Beyond Gaussian Assumptions: A General Fractional HJB Control Framework for Lévy-Driven Heavy-Tailed Channels in 6G), and allows physicists to elegantly track how the eigenvalues of massive random matrices evolve over time (Charles Bertucci et al. in A spectral dominance approach to large random matrices).
Key Foundational Work
Second‐order backward stochastic differential equations and fully nonlinear parabolic PDEs (2006) — contributed 3 concept elements to this direction.
Unifying Approaches and Removing Unrealistic Assumptions in Shape from Shading: Mathematics Can Help (2004) — contributed 1 concept elements to this direction.