A modified phase field approximation for mean curvature flow with conservation of the volume

• The classical conserved Allen–Cahn strategy adds a spatially uniform Lagrange multiplier to impose mass conservation, since mass approximates volume.The authors note that volume loss during numerical computations strongly affects dynamics, motivating better volume preservation.Source: §1 · pp. 4–5
• The paper proposes a phase-field model for mean-curvature flow with a forcing term and a volume constraint.The work also considers the approximation of forced mean-curvature motion.Source: §1 · pp. 2, 5
• The modified phase-field equation is rigorously proved to converge to the forced mean-curvature motion.The introduction identifies this as a proof following an argument due to.Source: §1 · p. 6
• The modified model has an O(ε²) volume error in the authors’ simulations, compared with O(ε) for the classical conserved Allen–Cahn equation.The introduction also reports better volume preservation for the modified model in numerical evidence.Source: §1 · p. 6

2 A modified reaction–diffusion equation for mean curvature flow with a forcing term

• The modified energy adds a forcing-dependent perturbation, Γ-converges to cW J, and its gradient flow is proposed to approximate the forced interface motion.The perturbation is continuous in the L1 topology, allowing the classical Modica–Mortola Γ-convergence result to extend to this energy.Source: §2 · pp. 6–7

2.1 Formal asymptotics for the modified Allen–Cahn equation

• For radial forcing and a unit-disk initial set, the formal expansion gives leading profile q, leading radius R, and vanishing first-order profile and radius corrections.The analysis treats a two-dimensional radial case; it also identifies the second-order profile correction U2, whose expression is not included in the supplied passages.Source: §2.1 · pp. 7, 9, 10

2.2 Proof of convergence for the modified phase field model

• The convergence proof places the phase-field interface within signed-distance O(ε|log ε|) of the forced mean-curvature flow under the theorem’s assumptions.The theorem gives an estimate for sufficiently small ε; the supplied passage states the interface-distance scale but omits the full estimate.Source: §2.2 · pp. 10, 14
• The proof applies the comparison principle and compares u_ε with sub- and supersolutions of (7).The passages state that a supersolution is used and that the comparison argument yields the corresponding bounds on u_ε.Source: §2.2 · p. 19
• A contradiction with (27) yields x ∈ Λ(t); the argument also establishes d(x,t) ≥ O(ε²|log ε|²), and proves (14).Source: §2.2 · pp. 19–20

3 Application to mean curvature flow with conservation of the volume

• The modified model's final term is related to mass conservation because the right-hand side has zero volume average.The term can also be understood as a Lagrange multiplier for the mass constraint.Source: §3 · p. 20
• The modified model has higher volume-conservation accuracy than the classical model, according to the section's formal analysis and numerical experiments.The authors state that the experiments support this claim, while noting that their arguments are formal.Source: §3 · pp. 20, 21, 23
• The formal comparison assumes the forcing terms depend on the respective phase-field solutions, and expects the stated asymptotic behavior if earlier results generalize despite that dependence.The authors explicitly qualify this expectation as an assumption.Source: §3 · p. 21
• For a regular bounded domain, the phase-field mass approximation differs from its volume by O(ε²).The estimate is retained on a fixed bounded set containing the domain when ε is sufficiently small.Source: §3 · p. 22
• The classical model is not expected to converge to the constrained motion at a rate better than O(ε), whereas the modified model is presented as more accurate.The section attributes the classical model's limit to a nonvanishing term of order ε.Source: §3 · p. 23

4 Numerical method and simulations

• The simulations use a smooth initial set strictly inside a fixed box Q and assume the evolving sets remain inside Q, allowing periodic boundary conditions.The computations impose a stability constraint for all choices of F.Source: §4 · p. 27, §4.1 · p. 27

4.1 Numerical scheme

• Equation (36) is approximated by splitting diffusion and reaction, treating diffusion in Fourier space and implementing that step with a fast Fourier transform.The first step has computational cost O(P^d log P).Source: §4.1 · p. 27

4.2 Numerical tests

• For an unforced shrinking circle, extinction-time error behaves like O(ε²|log ε|²), with numerical-scheme error negligible relative to modeling error for the chosen parameters.The test uses P = 28, Δt = 1/P², and ε = 2/P.Source: §4.2 · p. 28, Figure 3 · p. 29
• For a circle under constant forcing, both models give comparable extinction-time errors of O(ε²|log ε|²) for Cg = 2 and Cg = −2.The comparison is between phase field models (2) and (7).Source: §4.2 · pp. 28–29, Figure 4 · p. 29
• Extinction-time error scales as O(ε) for the classical model and O(ε²) for modified model (30) in the two-circle test.The test plots both radii over time and compares the models' extinction-time errors against ε.Source: §4.2 · p. 30, Figure 6 · p. 31
• With isotropic forcing g(x) = cg cos(8π|x|), changing cg significantly affects the classical-model evolution but has negligible impact on the modified model.The exact motion should remain stationary for every cg.Source: §4.2 · p. 31, Figure 7 · p. 31

5 Conclusion

• The modified phase-field model was rigorously proved to converge at the same O(ǫ2|log ǫ|2) order as classical Allen–Cahn.Source: §5 · p. 33
• The model approximates mean curvature flow with a forcing term.Source: §5 · p. 33
• O(ǫ2) volume error was observed for the modified model, versus O(ǫ) for the classical conserved Allen–Cahn equation in simulations.The model was formally derived for conserved mean curvature flow.Source: §5 · p. 33