In classical physics, we describe the world using partial differential equations (PDEs). These models are "local": the change in a system at a specific point depends only on its immediate neighbors. While local models like linear elasticity or the heat equation are the workhorses of engineering, they struggle with discontinuities—like a crack propagating through a wing or a sharp interface in a biological cell.
To handle these "broken" geometries, researchers use nonlocal models, such as peridynamics or nonlocal diffusion. Instead of derivatives, these models use integrals that sum up interactions over a small neighborhood called a horizon (denoted by $\delta$). However, a fundamental problem arises: as we make this horizon smaller and smaller, the nonlocal model should turn back into the classical local model. In the world of computer simulations, this transition is notoriously difficult to get right.
Asymptotic Compatibility (AC) is the mathematical bridge that ensures our numerical simulations remain physically consistent as they transition from the nonlocal world to the local one.
The Convergence Trap
When we simulate a nonlocal system, we have to deal with two different scales:
- The horizon ($\delta$), which defines the physical range of nonlocal interactions.
- The mesh size ($h$), which defines the resolution of our numerical grid.
In a perfect world, if we shrink both $\delta$ and $h$ to zero, our simulation should converge to the solution of the classical local PDE. But as Huaiqian You, Yue Yu, and Michael L. Parks point out in An asymptotically compatible meshfree quadrature rule for nonlocal problems with applications to peridynamics, most standard numerical schemes fail this test. They often require a very specific, computationally expensive ratio between $\delta$ and $h$ to stay stable. If you refine the mesh without shrinking the horizon correctly, the simulation might converge to the wrong answer—or not converge at all.
A scheme is asymptotically compatible if it converges to the correct local limit regardless of the path taken by $\delta$ and $h$ as they approach zero. This property allows engineers to refine their meshes freely without worrying about unphysical artifacts like "surface softening."
Building the Bridge: Optimization-Based Quadrature
The primary reason simulations lose consistency is poor integration. Nonlocal models require integrating over "interaction balls." When these balls are cut by a mesh or a boundary, standard integration rules (quadrature) fail.
To solve this, researchers like Yue Yu and Nathaniel Trask developed Optimization-Based Quadrature (OBQ). Instead of using fixed weights for integration, OBQ treats the weights as unknowns in a constrained optimization problem. The weights are chosen to perfectly reproduce certain polynomials (like constants, linear, and quadratic functions).
As shown in OBMeshfree: An optimization-based meshfree solver for nonlocal diffusion and peridynamics models, this "quadratic exactness" is the secret sauce. By forcing the discrete operator to behave correctly on simple polynomials, it automatically inherits asymptotic compatibility. This allows the solver to recover classical linear elasticity as the horizon vanishes, even in complex scenarios like linearly elastic composite materials.
Beyond Mechanics: Diffusion and Control
While peridynamics is the most famous application, the principle of asymptotic compatibility extends across physics:
- Diffusion and Phase Transitions: In Existence and local asymptotics for a system of cross-diffusion equations with nonlocal Cahn-Hilliard terms, Elisa Davoli and colleagues demonstrate that nonlocal models for mixing substances must rigorously converge to their local counterparts to be physically valid. Similarly, Kehan Shi shows that nonlocal evolution equations preserve the same decay rates as classical models, ensuring long-term stability.
- Boundary Conditions: One of the hardest parts of nonlocal modeling is applying forces to a surface. Huaiqian You and Yue Yu developed an asymptotically compatible treatment of traction loading, which uses geometric corrections to prevent the "ghost forces" that usually plague these interfaces.
- Optimization: Even in control theory, AC plays a role. Jan Friedrich and Claudia Nocita proved in Control of Conservation Laws in the Nonlocal-to-Local Limit that the best way to optimize a system with shocks (like traffic flow) is to use a nonlocal approximation that is—you guessed it—asymptotically compatible.
Why It Works: The Unified View
The intuition behind Asymptotic Compatibility is that nonlocality is a generalization, not a replacement. A good nonlocal discretization should contain the local model as a special case.
By using techniques like Reproducing Kernel Collocation (developed by Yu Leng, Xiaochuan Tian, and John T. Foster) or Quasinonlocal coupling (pioneered by Qiang Du), researchers have created a framework where we can "zoom in" on a fracture using nonlocal physics and "zoom out" to efficient local physics elsewhere, all while maintaining a single, mathematically consistent solution.
Asymptotic compatibility ensures that as our computational power grows and our grids get finer, our models don't just produce a solution—they produce the right physical solution.