Source-linked AI summary
Pointwise Monotonicity of the Allen--Cahn Flow and Dynamical Limitations of Energy-Stable Schemes
Pansheng Li, Dongling Wang
TL;DR
Energy dissipation does not by itself guarantee faithful pointwise dynamics, and spatial diffusion means the homogeneous ODE analysis does not determine the direction of inhomogeneous Allen–Cahn solutions. The paper establishes residual-based monotonicity benchmarks and tests energy-stable schemes against them. The exact flow and fully implicit Euler preserve the benchmark direction under stated conditions, while other energy-stable schemes can distort or reverse it.
Problem
Energy dissipation and maximum-bound preservation do not establish whether a discretization reproduces the local pointwise direction of spatially varying Allen–Cahn dynamics.
Source: §1 · pp. 2, 4Method
The paper defines sign-dependent admissible classes using the full Allen–Cahn residual and proves direction preservation for the exact flow and fully implicit Euler.
Source: §1 · p. 4Results
Takeaways & Limitations
Pointwise monotonicity provides a local dynamical criterion complementary to energy dissipation and maximum-bound preservation.
Source: §1 · p. 4The perturbation argument is local near δ = 0 and does not establish that δ = 1/4 lies within the guaranteed neighborhood.
Source: §5.2 · p. 16
Abstract
from arXiv · showhide
Energy stability is fundamental in the numerical approximation of phase-field models, but it does not by itself guarantee faithful reproduction of local dynamics. Recent ODE-level studies have shown that energy dissipation alone does not ensure dynamical fidelity for spatially homogeneous phase-field models (Xu and Xu, 2023; Li and Wang, 2026). However, these ODE arguments do not extend directly to spatially inhomogeneous Allen--Cahn solutions: diffusion enters the full residual $\mathcal R_\varepsilon(u):=Δu+\varepsilon^{-2}(u-u^3)$ and may change its pointwise sign even when $0<u\leq1$. Motivated by this distinction, we introduce sign-dependent admissible classes determined by the full Allen--Cahn residual and prove that the exact PDE flow preserves the corresponding pointwise monotone-growth or monotone-decay direction. The fully implicit Euler method inherits this structure in its standard unique-solvability regime. We then examine several widely used energy-stable schemes. First-order convex splitting and sufficiently strong first-order stabilization can preserve the correct direction, but on a reduced effective time scale, causing artificial delay or damping. By contrast, second-order convex splitting, stabilized CN/AB, IEQ, and SAV schemes admit monotone data for which large time steps generate wrong-signed pointwise increments despite dissipation of the original or a modified energy. For all four schemes, these strict reversals persist under sufficiently small smooth spatially nonhomogeneous admissible perturbations, so the counterexamples remain valid with genuinely active diffusion. Numerical experiments confirm this classification. Thus, pointwise monotonicity is a local dynamical criterion complementary to energy stability and maximum-bound preservation.
1. Introduction
Energy dissipation and range bounds do not determine the pointwise direction of spatially inhomogeneous Allen–Cahn dynamics. The paper defines residual-based monotonicity benchmarks and uses them to classify energy-stable schemes.
1. Introduction
- Energy stability constrains global energy dissipation, but does not guarantee the local pointwise direction of the dynamics.A scheme may remain within the maximum-bound range while still producing incorrect pointwise increments.Source: §1 · pp. 2, 4
- For sufficiently large spatial frequency, a profile with 0 < u0(x) < 1 has negative velocity at a sharp crest and positive velocity at a trough.Diffusion can overcome the positive reaction contribution near the crest, so range alone does not determine pointwise growth.Source: §1 · p. 3
- The exact flow preserves pointwise growth or decay for data whose full Allen–Cahn residual has the corresponding sign.The fully implicit Euler method inherits this structure in its unique-solvability regime.Source: §1 · p. 4
- First-order convex splitting and sufficiently strong first-order stabilization may preserve direction only on a distorted effective time scale, causing delay or damping.Second-order convex splitting, stabilized CN/AB, IEQ, and SAV can instead produce wrong-signed increments for large time steps despite energy dissipation.Source: §1 · p. 4
- For four schemes, strict sign reversals persist under sufficiently small smooth nonhomogeneous admissible perturbations with nonzero diffusion.One- and two-dimensional experiments with grid-refinement tests support the classification and indicate the reversals are not artifacts of an ODE reduction or spatial resolution.Source: §1 · p. 4
2. Pointwise Monotonicity of the Exact Allen–Cahn Flow
The full residual determines sign-dependent admissible data classes for which the exact Allen–Cahn flow preserves pointwise growth or decay. Sufficient bounds on the initial profile provide checkable examples of these classes.
2.1. Pointwise Monotonicity of the Allen–Cahn Flow
- The residual-sign condition defines positive- and negative-direction admissible classes, and the exact flow preserves their respective pointwise directions.For positive-class data, ut remains nonnegative; the negative-class result follows by the symmetry u → −u.Source: §2.1 · p. 5
- The monotonicity theorem uses the evolution equation for ut and the parabolic maximum principle to propagate the initial residual sign.The time derivative solves a linear parabolic equation with bounded coefficient on each finite time interval.Source: §2.1 · p. 5
2.2. Examples of admissible initial data
- If 0 < m ≤ u0 ≤ M < 1 and ε^2 ∥∆u0∥L∞ ≤ m(1 − M^2), then the exact solution satisfies ut ≥ 0.The section also gives constant, shifted-cosine, and small periodic perturbation examples of admissible initial data.Source: §2.1 · p. 5, §2.2 · p. 5
3. Fully Implicit Euler Scheme and Pointwise Monotonicity
Under the strict time-step condition τ < ε^2, fully implicit Euler preserves pointwise monotonicity when the discrete residual is nonnegative. Residual nonnegativity propagates from one step to the next.
3. Fully Implicit Euler Scheme and Pointwise Monotonicity
- The strict condition τ < ε^2 ensures strict convexity and unique solvability for the fully implicit step, supporting the monotonicity proof.The proof rules out a negative part of the increment using the residual sign and the boundary conditions.Source: §3 · p. 6
- For 0 < τ < ε^2, nonnegative discrete residual implies un+1 ≥ un pointwise, and residual nonnegativity propagates to the next step.Thus initially nonnegative residual yields a nondecreasing sequence at every spatial point.Source: §3 · p. 6
- The fully implicit Euler monotonicity theorem applies directly to initial data in the continuous positive-direction admissible class.The scheme's residual condition is compatible with that class.Source: §3 · p. 6
4. Dynamical Limitations of Unconditionally Energy-Stable Schemes
The section examines four mechanisms for unconditionally energy-stable or modified-energy-stable Allen–Cahn schemes. Its central distinction is that global energy dissipation alone does not ensure local pointwise monotonicity.
4. Dynamical Limitations of Unconditionally Energy-Stable Schemes
- The schemes considered use convex splitting, stabilization, invariant energy quadratization, and scalar auxiliary variables.They are representative mechanisms for unconditional energy stability or modified-energy stability.Source: §4 · p. 7
- Global energy dissipation does not, by itself, guarantee preservation of the local pointwise monotone-growth direction.The section evaluates the schemes against the pointwise direction established for the continuous flow.Source: §4 · p. 7
- The energy-stability analysis concerns either the original energy or a modified energy, while the dynamical criterion concerns local pointwise behavior.The section treats both unconditionally energy-stable and modified-energy-stable schemes.Source: §4 · p. 7
4.1. Convex splitting schemes
Convex splitting ranges from first-order schemes that preserve pointwise direction but distort physical time to a second-order scheme that can reverse the direction for large time steps despite modified-energy stability.
4.1.2. E
- CSS1 is unconditionally stable for the original energy and preserves pointwise direction by equaling fully implicit Euler at a reduced effective step.The effective step is τe = ε^2τ/(ε^2+τ), bounded above by ε^2.Source: §4.1 · p. 7, §4.1.1 · p. 7, §4.1.2 · p. 7
4.1.3. Large-step delay
- As τ grows, CSS1’s effective implicit-Euler evolution lags nominal physical time, so one large step does not reach the correct equilibrium.The scheme can remain monotone in step index while substantially delayed in physical time.Source: §4.1.3 · p. 8
4.1.4. Second-order convex splitting of Crank–Nicolson /Adams–Bashforth type
- For sufficiently large time steps, second-order CN/AB convex splitting reverses a monotone increment, and the strict reversal persists under small smooth nonhomogeneous perturbations.The counterexample applies at finite sufficiently large steps and can have genuinely active diffusion.Source: §4.1.4 · pp. 9–10
4.2. Stabilized semi-implicit schemes
The first-order stabilized scheme can preserve the residual-selected pointwise direction under strong stabilization, but this comes with time-scale distortion; the second-order stabilized CN/AB scheme can reverse that direction for large steps despite modified-energy stability.
4.2.2. Pointwise monotonicity of the first-order stabilized scheme
- With S ≥ 2/ε^2, the first-order stabilized scheme preserves nondecreasing pointwise evolution in [0, 1] from data with nonnegative full residual.Its increment is obtained by applying an elliptic resolvent to the residual.Source: §4.2.2 · p. 10
4.2.3. Artificial damping and large-step behavior of the first-order scheme
- The first-order stabilized update uses the reduced effective step τS = τ/(1 + Sτ), so large nominal steps produce artificial damping and fail to reach the correct equilibrium.Pointwise monotonicity for every τ therefore does not guarantee faithful physical-time relaxation.Source: §4.2.3 · p. 11
4.2.4. A second-order stabilized CN /AB scheme
- For sufficiently large τ, stabilized CN/AB reverses a pointwise increment even after an exact increasing start, and the strict reversal persists for small smooth nonhomogeneous perturbations.The construction also applies in the modified-energy-stable regime and remains valid with genuinely active diffusion.Source: §4.2.4 · p. 12
4.3. Invariant energy quadratization scheme
The first-order IEQ scheme dissipates a modified energy but can still reverse pointwise monotonicity for large time steps, including after small smooth perturbations that activate diffusion.
4.3. Invariant energy quadratization scheme
- IEQ’s unconditional modified-energy stability does not by itself control the sign of the pointwise phase increment.The counterexample’s first iterate exceeds 1, after which the reaction contribution has the opposite sign.Source: §4.3 · p. 13, §4.3.1 · p. 13
4.3.1. Large-step failure of pointwise monotonicity for IEQ
- For sufficiently large τ, the first-order IEQ scheme produces a strict decrement from admissible monotone data, so it is not unconditionally pointwise monotone.The strict inequality persists under sufficiently small smooth admissible perturbations with nonzero diffusion.Source: §4.3.1 · p. 13
- The homogeneous counterexample starts from a constant value in (0, 1), with the auxiliary variable initialized consistently as q0 = Q0.The phase iterates remain spatially homogeneous in this construction.Source: §4.3.1 · p. 13
4.4. Scalar auxiliary variable scheme
The first-order SAV scheme is unconditionally stable with respect to a modified energy, yet large time steps can reverse pointwise monotonicity; the strict reversal persists under small smooth nonhomogeneous perturbations.
4.4. Scalar auxiliary variable scheme
- SAV’s unconditional modified-energy stability does not by itself control the sign of the pointwise phase increment.The homogeneous construction initializes the scalar auxiliary variable with r0 = S0.Source: §4.4 · p. 14, §4.4.1 · p. 14
4.4.1. Large-step failure of pointwise monotonicity for SAV
- For sufficiently large τ, the first-order SAV scheme produces a strict decrement from admissible monotone data, so it is not unconditionally pointwise monotone.The sign reversal persists for sufficiently small smooth admissible perturbations with nonzero diffusion.Source: §4.4.1 · p. 14
- The SAV counterexample uses constant initial data in (0, 1), and the scheme preserves spatial homogeneity for those data.The strict homogeneous reversal is robust to small smooth admissible perturbations.Source: §4.4.1 · p. 14
5. Persistence under spatially nonhomogeneous admissible perturbations
The four counterexamples are first made explicit on the homogeneous invariant class, then shown to persist under sufficiently small smooth spatial perturbations, with diffusion genuinely active.
5.1. Second-order convex splitting
- Second-order convex splitting produces a strictly wrong-signed increment for genuinely nonhomogeneous admissible data, although the exact PDE trajectory is pointwise nondecreasing.Source: §5.1 · p. 15
5.2. Stabilized CN /AB scheme
- For stabilized CN/AB, a strict homogeneous decrement persists under sufficiently small smooth admissible perturbations, so the sign reversal remains when diffusion is active.Source: §5.2 · p. 16
5.3. IEQ scheme
- The IEQ scheme’s loss of pointwise monotonicity also occurs for genuinely nonhomogeneous admissible data.Source: §5.3 · p. 16
5.4. SAV scheme
- The SAV scheme’s strict sign reversal persists under sufficiently small nonhomogeneous admissible perturbations.Source: §5.4 · p. 17
6. Numerical experiments
The experiments show that energy stability and maximum-bound preservation do not ensure faithful pointwise dynamics: some schemes delay otherwise monotone evolution, while others reverse its direction at large time steps. Continuous-flow diagnostics and a fully implicit Euler computation provide monotone benchmarks for these comparisons.
6. Numerical experiments
- The experiments progress from initial-velocity diagnostics to scheme comparisons, including genuinely two-dimensional tests on two spatial grids.The two-dimensional comparisons use grids 1282 and 2562 to check whether increment patterns persist when both spatial directions are active.Source: §6 · p. 18
6.1. Initial-velocity diagnostic for the continuous problem
- Two positive profiles in the same range can have different initial PDE directions because their full residuals have different signs.The mixed-residual profile stays between 0 and 1 yet has negative velocity near crests and positive velocity near troughs.Source: §6.1 · p. 18
6.2.1. Common setting and diagnostics
- Scheme comparisons use a residual-positive cosine datum, and two-step tests start from a pair already in the positive monotone direction.A negative increment therefore records a reversal by the tested two-step update, rather than an initially mixed direction.Source: §6.2.1 · pp. 18–19
6.2.2. Reference computation: fully implicit Euler
- The fully implicit Euler reference remains pointwise nondecreasing in the tested convexity regime.With τ = 0.03 < ε2 = 0.25, its minimum increment is approximately 1.54 × 10−10 > 0.Source: §6.2.2 · p. 19, Figure 2 · p. 20
6.2.3. Monotone but delayed dynamics: CSS1 and first-order stabilization
- CSS1 and strongly stabilized first-order schemes retain the tested pointwise direction at large steps but exhibit delayed or damped transients.For stabilized semi-implicit evolution, the nominal-time error is ∥ustab(T) − uref(T)∥2 ≈ 9.510 × 10−1, while the damped-time error is ∥ustab(T) − uref(Teff)∥2 ≈ 1.373 × 10−2.Source: §6.2.3 · pp. 19–21
6.2.4. Two-step schemes: loss of pointwise monotonicity
- At large steps, second-order convex splitting overshoots and decreases, while stabilized CN/AB reverses pointwise direction even inside [0, 1].For convex splitting, Umax ≈ 1.188961 and m∆,l ≈ −1.87×10−1; for CN/AB, m∆,l ≈ −4.37 × 10−2 despite satisfying the modified-energy condition S = 5 ≥ 4.Source: §6.2.4 · p. 22
6.2.5. Auxiliary-variable schemes: IEQ and SAV
- Large-step IEQ and SAV runs develop negative pointwise increments despite remaining monotone in their smaller-step runs.The reported large steps are τ = 1, compared with τ = 0.1 for the monotone runs.Source: §6.2.5 · p. 23
6.3. Two-dimensional tests and grid refinement
The two-dimensional tests show that the sign classification persists on genuinely nonhomogeneous data and across a factor-two grid refinement. First-order methods preserve the tested direction, while several second-order and auxiliary-variable methods produce negative increments.
6.3. Two-dimensional tests and grid refinement
- The tests use genuinely two-dimensional periodic data with a positive discrete initial residual, and compare 128^2 and 256^2 grids.The nonlinear equations are iterated to tolerance 2 × 10^-11, with two-step schemes initialized by a monotone fully implicit Euler starter.Source: §6.3 · pp. 23–24
- 1.762821 × 10^-3 and 1.762877 × 10^-3 are the minimum increments for fully implicit Euler on the 128^2 and 256^2 grids, respectively.Both grids show Dh > 0.Source: §6.3 · p. 24, Figure 11 · p. 25
- 3.421144 × 10^-3 and 3.421247 × 10^-3 are the CSS1 minimum increments on the two grids, preserving pointwise increase despite the delayed transient seen in one-dimensional tests.Figure 12 reports a positive diagnostic with unchanged spatial pattern under refinement.Source: §6.3 · p. 24, Figure 12 · p. 25
- 2.473010 × 10^-2 and 2.472999 × 10^-2 are the minimum increments for first-order stabilization on the two grids, preserving the sign while retaining artificial damping.The diagnostic magnitude and shape are essentially identical under refinement.Source: §6.3 · p. 24, Figure 13 · p. 25
- For CS-modCN, the minimum increment is −1.866301×10^-1 on the 128^2 grid and −1.866300×10^-1 on the 256^2 grid.The negative increment is several orders of magnitude larger than the nonlinear tolerance and remains unchanged under refinement.Source: §6.3 · p. 24, Figure 14 · p. 26
- Stabilized CN/AB gives minimum increments of −4.283470 × 10^-2 and −4.283526 × 10^-2, with a negative region covering approximately 87.4% of grid points on both meshes.The location and geometry of this region remain stable under refinement.Source: §6.3 · p. 26, Figure 15 · p. 26
- IEQ minima are −1.031358 and −1.031322, while SAV minima are −2.962691 × 10^-1 and −2.962770 × 10^-1 on the two grids.For SAV, the negative region covers approximately 98.2% of grid points on both meshes; both methods retain their wrong-signed increments under refinement.Source: §6.3 · p. 26, Figure 16 · p. 27, Figure 17 · p. 27
6.4. Synthesis of the numerical evidence
- Across the two grids, positive increments persist for fully implicit Euler, CSS1, and first-order stabilization, while CS-modCN, stabilized CN/AB, IEQ, and SAV retain negative increments.The relative change in min Dh is below 3.5 × 10^-5 for every scheme; the synthesis also notes that stabilized CN/AB can remain within the maximum-bound interval while losing pointwise monotonicity.Source: §6.3 · p. 26, Table 1 · p. 27, §6.4 · pp. 27–28
7. Concluding remarks
The paper characterizes pointwise monotonicity through the sign of the full Allen–Cahn residual and shows when the exact flow and time discretizations preserve that direction. Its scheme comparisons distinguish directional preservation with time distortion from wrong-signed increments despite energy stability.
7. Concluding remarks
- The sign of the Allen–Cahn residual selects admissible data classes whose exact flow grows pointwise toward +1 or decays toward −1.This captures both where the solution lies and its direction of evolution at each spatial point.Source: §7 · p. 28
- Under the usual unique-solvability restriction and an appropriate discrete-residual sign, fully implicit Euler preserves the pointwise direction and propagates the residual sign.The paper gives a computable time-step condition for inheriting the monotone-growth mechanism.Source: §7 · p. 28
- First-order convex splitting and strong first-order stabilization can preserve direction with a reduced effective time scale, whereas second-order convex splitting, stabilized CN/AB, IEQ, and SAV can reverse it for sufficiently large steps despite energy dissipation.The negative statements are explicit counterexamples, not claims of failure for every initial datum.Source: §7 · p. 28
- For second-order convex splitting, stabilized CN/AB, IEQ, and SAV, strict homogeneous sign reversals persist under sufficiently small smooth nonhomogeneous admissible perturbations.The perturbations allow diffusion to participate genuinely in the PDE dynamics.Source: §7 · p. 28
- Energy stability, maximum-bound preservation, and pointwise monotonicity are distinct properties, so energy dissipation and range control do not determine the local time-increment sign.The paper recommends complementing energy or bound checks with local direction and physical-time-scale assessment when transient fidelity matters.Source: §7 · p. 28
Appendix A. Duhamel proof of the spatially homogeneous reduction
The appendix proves that spatially constant data remain spatially constant under periodic or homogeneous Neumann boundary conditions, reducing the flow on that invariant class to the scalar ODE. It also identifies why this reduction does not extend to nonzero constants under homogeneous Dirichlet conditions.
Appendix A. Duhamel proof of the spatially homogeneous reduction
- Spatially constant initial data form an invariant class under periodic or homogeneous Neumann boundary conditions, and the Allen–Cahn flow there is exactly the scalar ODE.The proof uses invariance of constants under the heat semigroup and uniqueness of the global mild solution.Source: Appendix A · p. 29
- The proof verifies that the spatially constant ODE solution satisfies the same Duhamel formula as the PDE solution, then invokes uniqueness.Vanishing diffusion at the initial time alone is not an invariance proof.Source: Appendix A · p. 29
- The reduction fails for nonzero constants under homogeneous Dirichlet conditions because constants are not in the Dirichlet Laplacian kernel; only zero is an exception.The Dirichlet heat semigroup immediately makes a nonzero constant spatially nonconstant.Source: Appendix A · p. 29