In continuum mechanics and physics, local models rely on partial differential equations (PDEs) to describe phenomena like diffusion or elasticity. These models assume that physical actions happen infinitesimally close to a point, using spatial derivatives to compute changes. However, when dealing with discontinuities like cracks or material fractures, derivatives break down because the underlying fields become non-differentiable.
To circumvent this, nonlocal models—such as peridynamics—replace derivatives with integral operators. Instead of looking only at an infinitesimal point, a nonlocal model integrates forces or fluxes over a finite neighborhood of radius $\delta$, known as the horizon.
When we simulate these systems numerically, we introduce a second spatial scale: the discretization mesh size or particle spacing, $h$. This sets up a delicate mathematical tension between two distinct limits:
- The numerical limit ($h \to 0$), where our discrete approximation should converge to the continuous nonlocal model.
- The physical limit ($\delta \to 0$), where the nonlocal model should recover classical local physics as the interaction range shrinks to zero.
Discrete Nonlocal Model (h, δ)
/ \
/ \
h → 0 (Mesh) / \ δ → 0 (Physical Horizon)
v v
Continuous Nonlocal Model Discrete Local Model
\ /
δ → 0 (Horizon)\ / h → 0 (Mesh)
v v
Classical Local PDE
If you refine the mesh ($h \to 0$) while keeping the horizon $\delta$ fixed, standard numerical schemes work well. But if you try to refine both simultaneously—shrinking the interaction scale alongside the mesh scale—standard methods frequently break down. They often converge to an incorrect physical limit, generating unphysical artifacts or "ghost forces."
The solution to this problem is Asymptotic Compatibility (AC): a design principle ensuring that a numerical scheme converges to the correct classical local limit regardless of the path $h$ and $\delta$ take as they approach zero.
The Commutativity Crisis
The root cause of numerical failure in multiscale simulations is that the limits $h \to 0$ and $\delta \to 0$ do not naturally commute. If a numerical scheme is not asymptotically compatible, the ratio of $h/\delta$ dictates the final solution. If $h$ shrinks faster than $\delta$, you get one answer; if they shrink proportionally, you might get a completely different, unphysical result.
As explained by [Yu Leng] et al. in Asymptotically Compatible Reproducing Kernel Collocation and Meshfree Integration for Nonlocal Diffusion, this lack of compatibility creates a severe disconnect. Practitioners cannot reliably use a single numerical framework to transition from highly nonlocal regimes to classical local regimes.
The underlying sensitivity of these systems is profound. As Bernard Deconinck demonstrated in Singular instability of exact stationary solutions of the non-local Gross–Pitaevskii equation, even an infinitesimally small amount of nonlocality can trigger rapid system breakdowns that local higher-order derivative approximations completely fail to predict. Because nonlocality introduces a singular gap between integral realities and differential approximations, the numerical integration scheme must be constructed with extreme care.
Why Standard Schemes Fail (and How AC Fixes Them)
Standard numerical methods typically approximate the nonlocal integral operator by using fixed geometric quadrature rules over the horizon volume. When the horizon $\delta$ shrinks, the number of discretization points falling within the horizon decreases. If $\delta$ approaches zero at a rate comparable to $h$, the quadrature error explodes because the discrete stencil can no longer accurately resolve the shape of the integral kernel.
To achieve asymptotic compatibility, the discrete operator must be forced to mimic the behavior of the local differential operator in the limit. Two primary mechanisms make this possible:
1. Optimization-Based Meshfree Weights
Instead of using rigid geometric rules to determine quadrature weights, modern AC schemes use optimization. As detailed by [Yiming Fan] et al. in An asymptotically compatible probabilistic collocation method for randomly heterogeneous nonlocal problems, these methods solve a local optimization problem at each particle to calculate weights that explicitly satisfy polynomial consistency conditions.
By forcing the discrete weights to exactly reproduce low-order polynomials, the discrete integral operator is mathematically guaranteed to morph into the correct local differential operator as $\delta \to 0$. [Huaiqian You] et al. implemented this mechanism in OBMeshfree: An Optimization-Based Meshfree Solver for Nonlocal Diffusion and Peridynamics Models, demonstrating that it successfully bridges nonlocal and classical continuum mechanics even in the presence of material heterogeneity.
2. Scaling and Parameter Alignment
Another approach relies on maintaining a strict, mathematically sound relationship between the discretization scale and the regularization scale. In Monotone Meshfree Methods for Linear Elliptic Equations in Non-divergence Form via Nonlocal Relaxation, Xiaochuan Tian showed that allowing the discretization parameter $h$ to be proportional to the nonlocal parameter $\delta$ provides stable, monotone convergence properties. This careful balancing prevents the stencil from degenerating as the nonlocality vanishes.
Resolving the Boundary Problem
The most difficult region to maintain asymptotic compatibility is near physical boundaries. For a point deep within a material, its interaction horizon is a perfect sphere. But for a point near the boundary, its horizon extends outside the domain, meaning it suffers from an incomplete interaction neighborhood.
Deep in Domain: Full Horizon Near Boundary: Truncated Horizon
. . . . . . . . |
. . .
* . . . . .
* | Boundary
. . . . . . . . |
This truncation causes artificial "surface effects" and makes applying standard boundary conditions (like forces or fluxes) highly error-prone. If handled incorrectly, the discrete model will fail to converge to classical boundary conditions as $\delta \to 0$.
To resolve this, [Yue Yu] et al. developed a reformulated framework in An asymptotically compatible treatment of traction loading in linearly elastic peridynamic fracture. By utilizing optimization-based quadrature alongside a novel volumetric treatment of traction forces, they ensured that the discrete boundary operators converge smoothly to classical elasticity solutions without boundary-induced integration errors.
Similarly, [Huaiqian You] et al. addressed flux conditions in An asymptotically compatible approach for Neumann-type boundary condition on nonlocal problems. They formulated a novel nonlocal Neumann boundary condition that achieves optimal second-order convergence to the classical local PDE counterpart, effectively eliminating artificial surface phenomena during multiscale transitions.
The Broader Impact
The mathematical foundation of asymptotic compatibility relies on unifying the function spaces of local and nonlocal calculus. [Qiang Du] et al. provided this bridge in Nonlocal trace spaces and extension results for nonlocal calculus by developing nonlocal trace spaces that remain fully consistent with classical local Sobolev spaces as the interaction range vanishes.
With these mathematical guarantees in place, engineers can confidently deploy nonlocal models in complex, large-scale engineering applications. For instance, [Masoud Behzadinasab] et al. demonstrated in Coupling of IGA and Peridynamics for Air-Blast Fluid-Structure Interaction Using an Immersed Approach that embedding nonlocal solid mechanics within smooth background meshes allows fracture and fragmentation patterns to converge to unique, mesh-independent solutions.
By ensuring that the underlying discrete math respects the local physical limits, asymptotically compatible schemes allow simulations to exploit the fracture-modeling power of nonlocality without losing the foundational correctness of classical continuum mechanics.
Key Foundational Work
Asymptotically compatible reproducing kernel collocation and meshfree integration for the peridynamic Navier equation (2020) — contributed 3 concept elements to this direction.
Singular instability of exact stationary solutions of the non-local Gross–Pitaevskii equation (2003) — contributed 1 concept elements to this direction.