---
title: Nonlocal-to-Local Limiting Strategy
url: https://www.emergentmind.com/topics/nonlocal-to-local-limiting-strategy
type: topic
---

# Nonlocal-to-Local Limiting Strategy

Nonlocal-to-local limiting strategy denotes a class of analytical and computational procedures in which a family of nonlocal models is shown to converge to a local partial differential equation, variational problem, or differential operator as the interaction kernel concentrates, the horizon vanishes, or a singular parameter approaches its local endpoint. In the cited literature, this passage is realized in several distinct regimes: convolution kernels are rescaled to a Dirac delta in nonlocal conservation laws and Cahn–Hilliard systems, the order parameter \(a\uparrow 2\) in nonsymmetric jump processes, the fractional parameter \(s\to 1^{-}\) in nonlocal minimal surfaces, the peridynamic horizon tends to \(0\) in linearized viscoelasticity and elasticity coupling, and graph-localized transport weights converge to a tensor-mobility continuity equation in Euclidean space [2304.01309] [1908.00945] [2203.07418] [1105.1158] [2402.16386] [2306.03475].

## 1. Representative limiting regimes

The underlying limit parameter varies by model class, but the objective is the same: identify a local object that retains the correct diffusion, drift, entropy, curvature, or mobility structure of the nonlocal system. In some settings the limit is a scalar conservation law, in others a Laplacian, a second-order divergence-form operator with drift, a local Kelvin–Voigt system, or a fourth-order aggregation–diffusion equation.

| Nonlocal setting | Limiting parameter | Local limit |
|---|---:|---|
| Nonlocal conservation laws with rescaled kernels | \(\varepsilon\to 0^+\) | \(\partial_t \rho+\partial_x(V(\rho)\rho)=0\) |
| Polygonally approximated nonlocal diffusion | \(\delta\to 0\) | \(\mathcal L_0=\Delta\) |
| Nonsymmetric nonlocal forms | \(a\uparrow 2\) | \(\partial_i(a_{ij}\partial_j u)+b_i\partial_i u\) |
| Nonlocal minimal surfaces | \(s\to 1^{-}\) | classical perimeter / mean curvature |
| Nonlocal viscous Cahn–Hilliard | \(\varepsilon\searrow 0\) | local Cahn–Hilliard with \(-\Delta u\) |
| Peridynamic Kelvin–Voigt viscoelasticity | horizon \(\to 0\) | \(-\nabla\cdot(D\varepsilon(\dot u)+C\varepsilon(u))=0\) |
| Localizing infinite graphs | \(\varepsilon\to 0\) | \(\partial_t\rho=\operatorname{div}\bigl(\rho\,\mathbb T(\nabla K*\rho+\nabla P)\bigr)\) |

For one-dimensional traffic-type conservation laws, the local limit is typically produced by a kernel family \(\eta_\varepsilon(x)=\varepsilon^{-1}\eta(x/\varepsilon)\) with \(\eta_\varepsilon\rightharpoonup \delta_0\); the limit equation is the scalar conservation law with flux \(V(u)u\) [2304.01309]. In nonlocal diffusion on \(\mathbb R^2\), the normalization \(\int_{B_\delta(0)} z_i^2 \widetilde\gamma_\delta(\|z\|_2)\,dz=1\) identifies the Laplacian as the local limit and makes asymptotic compatibility a moment condition rather than a purely geometric one [2109.12485]. For nonsymmetric forms, the scaling \((2-a)|x-y|^{-d-a}\) produces a second-order local energy, while the antisymmetric part survives as a first-order drift term [2203.07418].

A closely related pattern appears in short-range aggregation models and Gross–Pitaevskii equations. In the aggregation–diffusion setting, the scaling \(W(x)=-a^{-d}\varphi(x/a)\) and a Taylor expansion of the convolution produce a fourth-order thin-film/Cahn–Hilliard-type equation [2505.08443]. In the nonlocal Gross–Pitaevskii problem, a family \(W_\lambda\rightharpoonup \delta_0\) yields convergence of nonlocal dark solitons to the explicit local dark solitons of the one-dimensional Gross–Pitaevskii equation [2408.03870].

## 2. Structural devices that make the limit identifiable

A recurring feature of successful nonlocal-to-local arguments is the availability of a structural identity that converts the nonlocal quantity into an object with a local evolution equation. For the one-sided exponential kernel in traffic models, the “look-ahead” variable
\[
W[\rho](t,x)=\frac1\eta\int_x^\infty e^{(x-y)/\eta}\rho(t,y)\,dy
\]
satisfies
\[
\partial_x W=\frac1\eta W-\frac1\eta \rho,
\qquad\text{so in the normalized case}\qquad
\rho=W-\partial_x W.
\]
More importantly, \(W\) obeys a transport equation with a structured nonlocal source term. This closure property is the backbone of the Oleĭnik-type approach and is specific to the exponential kernel [2304.01309]. The same exponential structure underlies the lane-changing balance-law system, where
\[
\rho^i=W_\eta^i-\eta\,\partial_x W_\eta^i
\]
allows the system to be rewritten in terms of \(W_\eta\), which is the quantity for which uniform BV estimates are proved [2309.03866].

In nonlocal diffusion and elasticity, the corresponding structural ingredient is moment balance. The local limit depends on exact second-moment normalization, and the wrong geometry can destroy the limit even when the approximation looks visually close. In the polygonal-neighborhood problem, uniformly bounded side number yields
\[
\mathcal L_{\delta,n_\delta}q(x)\not\to \mathcal L_0 q(x),
\qquad q(x)=\|x\|_2^2,
\]
so the approximated model can converge to the wrong multiple of the Laplacian; convergence is recovered when the polygonal refinement improves with \(\delta\), for example when \(n_{\delta_k}\to\infty\) in the quasi-uniform inscribed case [2109.12485]. In peridynamic Kelvin–Voigt viscoelasticity, radial symmetry, kernel normalization, and concentration near zero identify the limit tensors \(C\) and \(D\) and make the nonlocal strain quadratic forms converge to the local expressions in \(\varepsilon(u)\) and \(\operatorname{div}u\) [2402.16386].

Operator-theoretic and graph limits use the same principle in a different language. For nonsymmetric jump kernels, the coefficients
\[
a_{ij}(x)=\lim_{a\uparrow2} a(2-a)\int_{B_1(0)} (-h_i)(-h_j)\,K^{(a)}(x,x+h)\,dh,
\qquad
b_i(x)=\lim_{a\uparrow2} a(2-a)\int_{B_1(0)} (-h_i)\,K^{(a)}(x,x+h)\,dh
\]
identify the diffusion matrix and the drift, respectively [2203.07418]. In the graph interaction equation, the limit tensor
\[
\mathbb T(x)=\frac12\,\widetilde\mu(x)\int_{\mathbb R^d} w\otimes w\,\vartheta(x,w)\,dw
\]
encodes the anisotropy of the localizing graph and is obtained from the localized edge weights by second-moment asymptotics [2306.03475]. These examples show that nonlocal-to-local limits are usually governed by a precise moment mechanism rather than by weak convergence alone.

## 3. Compactness, entropy, and admissibility

The central analytical obstacle is that weak convergence is usually insufficient to pass to the nonlinear flux or to identify the correct local solution. The conservation-law literature makes this point explicitly: total variation of \(u_\varepsilon\) itself may blow up, so the compact quantity is often the averaged field \(w_\varepsilon=u_\varepsilon*\eta_\varepsilon\), not the original state [2311.14528].

One major route is an Oleĭnik-type one-sided estimate. For the exponential-kernel traffic model, the nonlocal term \(W\) satisfies
\[
\frac{W(t,x)-W(t,y)}{x-y}\ge -\frac1{\kappa t},
\]
and under stronger structural assumptions the quantity \(g=V'(W)W\partial_x W\) satisfies a one-sided bound of the form
\[
\sup_{\mathbb R} g(t,\cdot)\le \frac{\|\rho_0\|_{L^\infty}}{\kappa t}.
\]
These estimates control oscillations, generate BV regularization for transformed quantities, yield precompactness in \(L^1_{\mathrm{loc}}\), and exclude non-entropic shocks in the limit. A notable feature is that the convergence to the local entropy solution is proved under the sole assumption \(\rho_0\in L^\infty(\mathbb R;\mathbb R_{\ge0})\), without assuming \(\mathrm{TV}(\rho_0)<\infty\) [2304.01309].

A second route is compensated compactness. For the class
\[
\partial_t u_\varepsilon+\partial_x\!\bigl(V(u_\varepsilon*\gamma_\varepsilon)\,u_\varepsilon\bigr)=0,
\]
strong \(L^1_{\mathrm{loc}}\)-convergence is obtained without total variation bounds or Oleĭnik estimates by proving compactness of the entropy productions
\[
\partial_t\eta(w_\varepsilon)+\partial_x q(w_\varepsilon)
\]
in \(W^{-1,2}_{\mathrm{loc}}\) and then applying Tartar–Murat compensated compactness. The two settings treated are the piecewise constant kernel \(\gamma=\mathbf 1_{[-1,0]}\) with Greenshields velocity \(V(\xi)=1-\xi\), and strictly monotone kernels with decreasing velocities, including the exponential kernel [2511.15631]. In the anisotropic traffic overview, a different compactness estimate is proved: if \(\eta\) is supported in \(\mathbb R_-\), nondecreasing, and convex on \(\mathbb R_-\), then
\[
\operatorname{TotVar} w_\varepsilon(t,\cdot)\le \operatorname{TotVar} w_\varepsilon(0,\cdot)
\]
for \(w_\varepsilon=u_\varepsilon*\eta_\varepsilon\), which again shifts compactness from \(u_\varepsilon\) to its nonlocal average [2311.14528].

Variational limits can also encode admissibility and boundary information. In the nonlocal viscous Cahn–Hilliard equation, the nonlocal operator \(B_\varepsilon\) converges to \(-\Delta\) in the variational sense, and the boundary condition \(\partial_{\mathbf n}u=0\) is not imposed at the nonlocal level but emerges from the \(\Gamma\)-limit of the energies [1908.00945]. In the control problem for conservation laws, the improved nonlocal-to-local convergence theorem allows simultaneous convergence of the kernels and the initial data, which is the key step in the subsequent \(\Gamma\)-convergence of the objective functionals [2510.00677].

## 4. Variational, evolutionary, and operator-theoretic formulations

A large part of the modern theory formulates the nonlocal-to-local passage at the level of energies, forms, semigroups, or gradient structures rather than only at the level of weak solutions. In nonsymmetric nonlocal forms, the convergence is proved in the sense of Mosco–Hino: for \(a_n\uparrow2\), the forms \((\mathcal E^{(a_n)},H^{a_n/2}(B))\) converge to \((\mathcal E,H^1(B))\), which implies convergence of resolvents, semigroups, and dual semigroups. The limiting local form is
\[
\mathcal E(u,v)=\int_B a_{ij}(x)\partial_i u\,\partial_j v\,dx
+2\int_B b_i(x)\partial_i u\,v\,dx,
\]
making precise the statement that antisymmetry behaves like a lower-order drift [2203.07418].

A broader abstraction is provided by nonlocal \(H\)-convergence on a closed exact complex \((A_0,A_1)\). There the coefficients are bounded linear operators \(a\in L(H_1)\), not necessarily multiplication operators, and convergence is defined through the solution operators of two dual elliptic problems. The theory proves uniqueness of the nonlocal \(H\)-limit, relative compactness in the topology \(T_H\), a block-matrix characterization through \(a_{00}\), \(a_{10}a_{00}^{-1}\), \(a_{00}a_{01}\), and the Schur complement, and the coincidence of local and nonlocal \(H\)-convergence for multiplication operators on the standard \((\nabla,\operatorname{curl})\) complex [1804.02026].

Gradient-flow formulations provide the evolutionary analogue of these static convergence principles. In linearized viscoelasticity, the quasistatic nonlocal Kelvin–Voigt evolution is written as
\[
dD_\rho(\dot u_\rho)+dE_\rho(u_\rho)=0
\]
and studied through the exact energy–dissipation balance
\[
E_\rho(u_\rho(t))+\int_0^t D_\rho(\dot u_\rho)\,ds
+\int_0^t D_\rho^*\!\bigl(-dE_\rho(u_\rho)\bigr)\,ds
=E_\rho(u_{\rho,0}).
\]
Following an evolutionary \(\Gamma\)-convergence argument, the nonlocal solutions converge strongly in \(C([0,T];L^2)\) to the unique weak solution of the local Kelvin–Voigt problem [2402.16386]. The graph interaction equation uses the same philosophy in a different geometry: graph dynamics are Finslerian gradient flows, the continuum limit is a weighted Wasserstein gradient flow with mobility tensor \(\mathbb T\), and the passage is formulated as EDP convergence of gradient structures [2306.03475].

Optimization problems also fit naturally into this framework. In the basis-pursuit limit for optimal design, the antisymmetric nonlocal mixed-norm problem provides only a one-sided estimate toward the local measure-valued limit, whereas dropping antisymmetry yields genuine \(\Gamma\)-convergence to the relaxed local dual problem [2107.12994]. In the control of conservation laws, the objective functionals
\[
\mathcal G_H(u_o)=\mathcal J(u_H(T))+\mathcal K(u_H)+\mathcal I(u_o)
\]
\(\Gamma\)-converge to the corresponding local functional, and minimizers of the nonlocal problems converge, up to subsequences, to minimizers of the local problem [2510.00677].

## 5. Boundary conditions, coupling, and asymptotic compatibility

A major computational manifestation of the nonlocal-to-local strategy is the construction of boundary or interface conditions that remain correct as the horizon shrinks. One approach converts local surface data into nonlocal volume constraints. In the nonlocal diffusion problem based on the nonlocal gradient/divergence calculus, the procedure is: solve a local Poisson problem with the available surface data, define the nonlocal Neumann data by
\[
\widetilde g_n(\mathbf x)=-\mathcal N(G u_l)(\mathbf x)\qquad \mathbf x\in\Omega_I,
\]
and then solve the nonlocal problem with those generated volume constraints. The resulting nonlocal solution \(w_n\) satisfies the quadratic estimates
\[
|||w_n-u_l||| \le C\,\varepsilon^2 \|D^{(4)}u_l\|_{\infty,\Omega},
\qquad
\|w_n-u_l\|_{0,\Omega}\le \widehat C\,\varepsilon^2\|D^{(4)}u_l\|_{\infty,\Omega},
\]
so the vanishing-horizon limit recovers the local solution in both the energy seminorm and \(L^2\) [1906.04259].

A second approach is local-to-nonlocal coupling by optimization. In the peridynamic–FEM setting, a static Linear Peridynamic Solid model is coupled with classical Navier–Cauchy elasticity on overlapping subdomains by minimizing
\[
J(\mathbf u_n,\mathbf u_l)=\frac12\int_{\Omega_o} |\mathbf u_n-\mathbf u_l|^2\,d\mathbf x
\]
subject to the two state equations and the virtual control data on the interface. The method relies on the facts that \(L_{\rm LPS}\to L_{\rm NC}\) as \(\delta\to0\) and that the LPS and NC equations are exactly equivalent for quadratic displacement fields [2110.04420].

The term asymptotically compatible is used explicitly in the local-to-nonlocal heat coupling literature. In the nonoverlapping Robin–Dirichlet coupling problem, a nonlocal Neumann-type boundary treatment is constructed by converting local flux into a collar-volume correction, then extended to a nonlocal Robin-type condition. The continuous formulation recovers the local Neumann problem with \(O(\delta^2)\) accuracy in \(L^\infty\), while the fully discrete coupled scheme shows asymptotic convergence to the local limit with observed \(O(\delta)=O(h)\) rate when \(\delta/h\) is fixed [1912.06270]. In one-dimensional diffusion, the quasinonlocal coupling method based on geometric reconstruction gives a variationally consistent mixed operator, passes the patch test, preserves flux balance and the maximum principle, and satisfies
\[
\|u_\delta^{\rm qnl}-u_0\|_{L^\infty(\Omega)}=O(\delta)
\]
as the interaction horizon shrinks [1704.00348].

These results also delimit what asymptotic compatibility does not permit. A polygonal approximation of a ball-shaped interaction neighborhood is not automatically safe in the vanishing-horizon regime: if the number of sides stays uniformly bounded, the operator can converge to the wrong local limit; geometric refinement must therefore accompany horizon refinement [2109.12485].

## 6. Scope, limitations, and open directions

The nonlocal-to-local limiting strategy is used across conservation laws, diffusion, viscoelasticity, minimal-surface theory, optimal design, graph transport, aggregation–diffusion, and dispersive wave problems. In traffic applications, it has been used for single-lane conservation laws, nonlocal lane-changing balance laws, and optimal control with the initial datum as control [2304.01309] [2309.03866] [2510.00677]. In geometry, it provides the bridge from \(s\)-minimal surfaces to classical minimal surfaces: as \(s\to1^{-}\), \((1-s)\mathcal J_s\) converges to the classical perimeter and the normalized nonlocal curvature converges to mean curvature, enabling an improvement-of-flatness theorem uniform in \(s\) near \(1\) [1105.1158]. In dispersive problems, it yields convergence of nonlocal gray and black solitons to the explicit local Gross–Pitaevskii dark solitons [2408.03870].

Several limitations are structural rather than technical. The local limit can be wrong if geometric fidelity is too coarse, as in polygonal interaction neighborhoods with bounded side number [2109.12485]. In basis-pursuit optimal design, antisymmetry is physically natural and mathematically useful for compactness, but it weakens the nonlocal-to-local passage to a one-sided estimate; genuine \(\Gamma\)-convergence requires dropping the antisymmetry constraint [2107.12994]. In anisotropic conservation laws, the compactness mechanism depends critically on one-sided kernels and the sign structure of \(\eta'\); symmetric or nonconvex kernels need different arguments [2311.14528]. These examples show that the local limit is not a universal consequence of “small horizon” alone.

Open problems remain model-dependent. In the local fourth-order aggregation–diffusion model obtained from short-range adhesion, the cited work emphasizes unresolved questions on uniqueness of weak solutions, existence of minimizers on the whole space, long-time asymptotics, and the extent of validity of the local approximation for real co-culture experiments [2505.08443]. A plausible implication is that future progress will continue to depend on choosing the transformed quantity, variational structure, or interface formulation that makes the small-horizon limit rigid enough to identify the correct local object. Across the cited literature, that choice is the defining content of the nonlocal-to-local limiting strategy.

Source: https://www.emergentmind.com/topics/nonlocal-to-local-limiting-strategy