Source-linked AI summary
Maximum principle preserving exponential time differencing schemes for the nonlocal Allen-Cahn equation
Qiang Du, Lili Ju, Xiao Li, Zhonghua Qiao
TL;DR
The paper addresses numerical approximation of the nonlocal Allen-Cahn equation while preserving its maximum-principle and energy properties. It combines quadrature-based finite differences with stabilized first- and second-order ETD schemes, and proves unconditional discrete maximum-principle preservation, error estimates, asymptotic compatibility, and discrete energy stability. The main scope boundary is that maximum-principle-preserving schemes are developed only through second order, while higher-order preservation remains open.
Problem
Numerical schemes for the NAC equation must address preservation of its maximum principle and energy dissipation at the discrete level.
Method
The paper combines quadrature-based finite differences for spatial discretization with stabilized first- and second-order ETD time integration.
Results
The proposed schemes unconditionally preserve the discrete maximum principle, admit maximum-norm error estimates, are asymptotically compatible, and are discretely energy stable.
Takeaways & Limitations
The schemes provide numerical approximations that retain key NAC properties while converging to the classic Allen-Cahn solution as δ, h, and τ vanish.
Takeaways & Limitations
Maximum-principle-preserving schemes are developed only through second order, and whether higher-order schemes can preserve the principle remains open.
Abstract
from arXiv · showhide
The nonlocal Allen-Cahn (NAC) equation is a generalization of the classic Allen-Cahn equation by replacing the Laplacian with a parameterized nonlocal diffusion operator, and satisfies the maximum principle as its local counterpart. In this paper, we develop and analyze first and second order exponential time differencing (ETD) schemes for solving the NAC equation, which unconditionally preserve the discrete maximum principle. The fully discrete numerical schemes are obtained by applying the stabilized ETD approximations for time integration with the quadrature-based finite difference discretization in space. We derive their respective optimal maximum-norm error estimates and further show that the proposed schemes are asymptotically compatible, i.e., the approximate solutions always converge to the classic Allen-Cahn solution when the horizon, the spatial mesh size and the time step size go to zero. We also prove that the schemes are energy stable in the discrete sense. Various experiments are performed to verify these theoretical results and to investigate numerically the relation between the discontinuities and the nonlocal parameters.
1. Introduction.
The paper develops maximum-principle-preserving and energy-stable numerical schemes for the nonlocal Allen-Cahn equation, whose nonlocal operator approaches the Laplacian as the interaction horizon vanishes.
- Problem setting: The NAC equation replaces the Laplacian with a horizon-parameterized nonlocal diffusion operator and models phenomena including phase transitions, peridynamics, image analysis, and heat conduction.The horizon δ measures the range of nonlocal interactions, while kernel assumptions enforce consistency with the standard Laplacian as δ → 0.
- Structural properties: The NAC equation satisfies a maximum principle, so bounded initial and boundary data produce solutions bounded by 1.This property parallels the local Allen-Cahn equation and motivates preserving boundedness after discretization.
- Structural properties: As an L2 gradient flow, the NAC solution decreases its energy over time according to an energy dissipation law.The paper treats preservation of this decay as a key issue for numerical stability.
- Paper objective: The work targets first- and second-order ETD schemes that preserve the discrete maximum principle, provide error estimates, achieve asymptotic compatibility, and remain discretely energy stable.The paper also reports numerical experiments examining these properties and the relation between discontinuities and nonlocal parameters.
- Prior numerical approaches: The paper uses exponential time differencing because it evaluates the linear contribution exactly, supporting stability and accuracy for stiff linear terms.The schemes remain explicit and efficient while applying suitable linear splitting techniques.
2. Fully discrete exponential time differencing schemes.
The fully discrete methods combine quadrature-based finite differences for nonlocal diffusion with stabilized ETD time stepping, including first- and second-order variants and fast Fourier implementations.
- 2.1. Quadrature-based finite difference semi-discretization.: The spatial discretization uses a quadrature-based finite difference approximation of the nonlocal diffusion operator on a uniform mesh.The resulting discrete operator is self-adjoint and negative semi-definite, with O(h2) consistency for fixed δ.
- 2.1. Quadrature-based finite difference semi-discretization.: A stabilizing parameter κ defines Lh := −ε2Dh + κIdN and f(U) := (κ + 1)U − U^3, yielding a positive definite linear operator.This splitting rewrites the semi-discrete system in a form suitable for ETD integration.
- 2.2. Exponential time differencing schemes for time-stepping.: ETD1 approximates the nonlinear term by its value at the beginning of each time step and evaluates the resulting temporal integral exactly.The ETD construction integrates the linear part exactly and explicitly approximates the nonlinear contribution.
- 2.2. Exponential time differencing schemes for time-stepping.: ETDRK2 uses linear interpolation between nonlinear evaluations at the current state and an intermediate approximation of the next state.The method provides the second-order ETD Runge-Kutta time-stepping variant.
- 2.3. Efficient implementations of the ETD schemes.: For periodic boundaries, matrix-function actions φγ(Lhτ) can be implemented with multidimensional FFTs at O(N^d log N) complexity per time step.The discrete Fourier transform diagonalizes the relevant operator action mode by mode.
3. Discrete maximum principle.
The paper establishes that its ETD1 and ETDRK2 schemes preserve the discrete maximum principle for arbitrary time-step sizes when the stabilizing parameter satisfies κ ≥ 2.
- Proof structure: The analysis is developed from matrix properties including negative diagonal entries and weak diagonal dominance.The nonlinear mapping is treated componentwise because it consists of independent scalar functions.
- Supporting lemma: The matrix estimate ∥e^−L_hτ∥∞ ≤ e^−κτ is a key ingredient in proving the discrete maximum principle.The proof uses the diagonal-dominance properties of the nonlocal stiffness matrix.
- Boundary conditions: The analysis is initially restricted to periodic boundary conditions, but the key matrix estimate also applies to Dirichlet boundary conditions.The corresponding Dirichlet stiffness matrix remains weakly diagonally dominant with negative diagonal entries.
- ETD1 scheme: For initial data satisfying ∥u_0∥L∞ ≤ 1, the ETD1 scheme preserves the discrete maximum principle for any τ > 0 when κ ≥ 2.The result is proved by induction on the time level.
- ETDRK2 scheme: For initial data satisfying ∥u_0∥L∞ ≤ 1, the ETDRK2 scheme preserves the discrete maximum principle for any τ > 0 when κ ≥ 2.Its proof controls the intermediate ETDRK2 value before applying the second stage.
4. Error estimates and asymptotic compatibility.
The paper analyzes convergence of ETD1 and ETDRK2 schemes both to the NAC solution at fixed horizon and to the local Allen-Cahn solution in the asymptotic-compatibility limit.
- Convergence to NAC: The numerical solutions converge to the exact NAC solution as h and τ go to zero for any fixed δ > 0.The paper derives maximum-norm error estimates under regularity assumptions on the exact solution.
- ETD1 error estimate: The ETD1 error analysis uses Lipschitz continuity of the nonlinear mapping and consistency of the spatial discretization.The estimates assume sufficient smoothness of the exact solution and use a discrete error evolution followed by Gronwall’s inequality.
- ETDRK2 error estimate: The ETDRK2 analysis establishes analogous maximum-norm error estimates using its intermediate-stage evolution and interpolation estimates.The constants are stated to be independent of δ, h, and τ under the listed regularity conditions.
- Asymptotic compatibility: The numerical solutions converge to the local Allen-Cahn solution as δ, h, and τ approach zero, which is the paper’s asymptotic compatibility result.The result is obtained by combining consistency of the nonlocal operator with analogous error analysis for the local problem.
- Asymptotic compatibility: The asymptotic-compatibility bounds have constants independent of δ, h, and τ under the stated smoothness assumptions.The result is stated for both ETD1 and ETDRK2, with the ETDRK2 time-step range restricted to τ ∈ (0, 1].
5. Discrete energy stability.
The paper establishes discrete energy stability for both proposed ETD schemes: ETD1 is unconditionally energy stable, while ETDRK2 has uniformly bounded discrete energy under stated conditions.
- ETD1 inherits the energy decay law in the discrete sense for the discretized energy E_h.
- The ETD1 scheme is unconditionally energy stable for any τ > 0.
- E_h(U^{n+1}) − E_h(U^n) ≤ (U^{n+1} − U^n)^T B_1(U^{n+1} − U^n) ≤ 0.The result follows from the negative definiteness of B_1.
- For ETDRK2, the analysis proves uniform boundedness of the discretized energy E_h.
- The ETDRK2 energy bound holds for any h > 0 and 1 ≥ τ > 0, with a constant independent of h and τ.
- The ETDRK2 proof uses the same energy-increment calculation as ETD1, together with positive-definite B_2 and the maximum-principle bounds when κ ≥ 2.
6. Numerical experiments.
Numerical experiments verify the ETD schemes’ temporal and spatial convergence, convergence toward the local Allen–Cahn limit, maximum-principle and energy-stability behavior, and the dependence of interfaces and discontinuities on the horizon parameter.
- Experimental setup: The experiments use fractional-power kernels with integrable and non-integrable regimes, and apply ETDRK2 broadly while reserving ETD1 for temporal convergence tests.All simulations use κ = 2 on a two-dimensional domain; the convergence tests begin from smooth data, while stability tests use random initial states.
- Convergence tests: Temporal maximum-norm errors exhibit the expected first-order ETD1 and second-order ETDRK2 convergence for both kernel regimes.The observed errors are almost independent of the choices of δ and α.
- Convergence tests: Spatial maximum-norm errors converge at almost second order for both integrable and non-integrable kernels, consistent with the theoretical estimates.The tests fix δ = 2 and use the ETDRK2 solution with N = 4096 as the benchmark.
- Convergence tests: The numerical NAC solutions converge to the LAC solution with observed second-order convergence as δ approaches zero.The comparison uses fixed N = 4096 and τ = T with ETDRK2.
- Stability tests: For δ = 3ε, NAC dynamics resemble LAC dynamics, while δ = 4ε produces thinner interfaces, discontinuities, and much slower evolution.The discrete maximum principle is preserved and discrete energy decays monotonically in all tested cases; δ = 4ε lies in the discontinuous regime because ε^2Cδ < 1.
- Discontinuity in the steady state solution: In bubble simulations, δ = 0.2 yields rapid shrinkage and disappearance, whereas δ = 0.8 and 3.2 produce sharpening interfaces, discontinuities, and steady states with expected jumps.The numerically computed steady-state jumps match the theoretical values very well.
7. Conclusions.
The paper develops maximum-principle-preserving ETD schemes for the nonlocal Allen–Cahn equation and establishes error, energy-stability, and asymptotic-compatibility results. The study is limited to schemes up to second order in time, while higher-order maximum-principle preservation remains open.
- The schemes combine quadrature-based finite differences in space with first-order ETD and second-order ETD Runge–Kutta time integration.
- The analyses derive error estimates and prove discrete energy stability and asymptotic compatibility for both schemes.
- Numerical experiments verify the theoretical results and examine solution properties associated with nonlocality.
- Higher-order schemes that preserve the maximum principle remain an open problem for future work.