A weighted scalar auxiliary variable method for solving gradient flows: bridging the nonlinear energy-based and Lagrange multiplier approaches

Cheng Yuan; Jerry Zhijian Yang; Qiong-Ao Huang; Wei Jiang

REVIEW 3 major objections 4 minor 42 references

A weighted scalar auxiliary variable method for solving gradient flows: bridging the nonlinear energy-based and Lagrange multiplier approaches

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The two main scalar auxiliary variable (SAV) approaches for gradient flows—one that always has solutions but controls only a modified energy, the other that preserves the original energy but can fail at large time steps—are unified by a…

desk verdict The weighted-SAV idea is solid and the numerics are convincing, but Theorem 3's existence guarantee is false as stated—a bounded-E_N counterexample with cos nonlinearity shows the proof's f_min bound fails. read the letter →

arxiv 2411.17403 v1 pith:GZD6VTLU submitted 2024-11-26 math.NA cs.NAphysics.comp-ph

Qiong-Ao Huang , Wei Jiang , Jerry Zhijian Yang , Cheng Yuan This is my paper · ORCID

classification math.NAcs.NAphysics.comp-ph MSC 65M2265M7035K3535K55

keywords weightedscalarauxiliaryvariablegradientflowenergystabilityLagrangemultiplierphase-fieldmodelCahn–HilliardequationAllen–Cahnsolutionexistence

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces a weighted scalar auxiliary variable (SAV) method for gradient flows, blending two existing SAV strategies: the nonlinear energy-based approach, which always has solutions but only guarantees stability for a modified energy, and the Lagrange multiplier approach, which preserves the original energy but can fail for large time steps. The central claim is that a weight coefficient $\lambda \in [0,1]$ interpolating between the two produces schemes that are solvable for any time step size whenever $\lambda$ is large enough, while choosing $\lambda$ as small as possible keeps the discrete energy close to the original free energy. First- and second-order discretizations are analyzed, and numerical experiments on Allen–Cahn, Cahn–Hilliard, and space-fractional Cahn–Hilliard equations demonstrate the expected accuracy, energy stability, and improved faithfulness of the energy. A reader should care because the method addresses a known trade-off in structure-preserving simulation between solution existence at large time steps and accurate energy dissipation.

What carries the argument

The central object is the weighted scalar auxiliary variable reconstruction (3.1), where $r(t)=\sqrt{E_N[\varphi]+C}$, $L$ is the linear, symmetric, non-negative operator in the quadratic energy, and $\lambda\in[0,1]$ is the weight; $\lambda=1$ reproduces the nonlinear energy-based SAV scheme and $\lambda=0$ reproduces the Lagrange multiplier formulation. The load-bearing mechanism is the decoupling $\varphi^{n+1}=p^{n+1}+r^{n+1}q^{n+1}$, which reduces each time step to two linear elliptic solves with constant coefficients plus one nonlinear scalar equation (3.7) for $r^{n+1}$. The positivity identity (3.11) makes the quadratic coefficient of that scalar equation positive for any $\tau>0$, and the assumed uniform boundedness of $E_N$ supplies the extremal argument in Theorem 3. The bisection algorithm then chooses the minimal $\lambda$ for which the scalar equation has a solution, and the weighted energy (3.14) is what is guaranteed to dissipate; original-energy dissipation follows when the inequality connecting $2r^{n+1}(r^{n+1}-r^n)$ and $E_N[\varphi^{n+1}]-E_N[\varphi^n]$ holds.

What would settle it

Run the first-order scheme (3.3) on the Cahn–Hilliard equation with an untruncated quartic $E_N$ and a large time step, following Algorithm 1; if at some step no $\lambda\in[0,1]$ makes the scalar equation (3.7) solvable, then Theorem 3 fails for the untruncated energy. A more direct check is to exhibit a configuration where the quadratic coefficient in (3.8) is positive but $f(x;\lambda)+g(x;\lambda)$ stays positive for all real $x$ at every $\lambda\in[0,1]$, contradicting the claimed solution existence.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that the nonlinear energy-based SAV method and the Lagrange multiplier SAV method are two ends of a single spectrum, and that the spectrum removes each method's defect. With auxiliary variable $r(t)=\sqrt{E_N[\varphi]+C}$, the weighted scheme (3.3) couples the update of $r^{n+1}$ to the actual change in the nonlinear energy $E_N[\varphi^{n+1}]-E_N[\varphi^n]$ through the weight $\lambda$. Theorem 3 states that for any time step $\tau>0$ there exists $\bar\lambda\in(0,1)$ such that for all $\lambda\in(\bar\lambda,1]$ the resulting nonlinear scalar equation (3.7) has a solution, so the scheme is solvable for arbitrary time steps. Theorem 4 proves unconditional dissipation of the weighted energy $E[\varphi,r] = \tfrac12(\varphi,L\varphi)+\lambda(r^2-C)+(1-\lambda)E_N[\varphi]$, and shows that when $2r^{n+1}(r^{n+1}-r^n)\ge E_N[\varphi^{n+1}]-E_N[\varphi^n]$ the original energy dissipates as well. By bisection on $\lambda$, the method selects the smallest weight for which the scalar equation is solvable at each step, numerically yielding an energy close to the original while avoiding the nonexistence failure of the pure Lagrange multiplier approach.

Load-bearing premise

The solvability theorem assumes the nonlinear part of the free energy, $E_N$, is uniformly bounded; the paper enforces this by truncating the double-well potential in all numerical tests, so the guarantee does not directly cover the standard untruncated quartic phase-field energy.

Editorial extensions

If this is right

First-order backward Euler and second-order Crank–Nicolson weighted SAV schemes retain their temporal convergence orders while permitting arbitrarily large time steps once the weight is chosen sufficiently large.
The adaptive minimal-weight strategy makes the computed discrete energy approximate the original free energy more closely than the pure nonlinear energy-based SAV scheme, as measured by energy deviation in the numerical experiments.
When $2r^{n+1}(r^{n+1}-r^n)\ge E_N[\varphi^{n+1}]-E_N[\varphi^n]$, the scheme dissipates the original energy, not merely a modified energy.
The computational cost stays consistent with classical SAV methods: two constant-coefficient linear elliptic solves per step plus one scalar root-finding problem.
The method covers $L^2$, $H^{-1}$, and fractional $H^{-\nu}$ gradient flows, including Allen–Cahn, Cahn–Hilliard, and space-fractional Cahn–Hilliard equations, and preserves mass conservation for the Cahn–Hilliard family.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Because the existence proof relies on the uniform bound $|E_N|\le M$, the theoretical guarantee applies to the truncated double-well potential used in the numerics; extending Theorem 3 to the untruncated quartic phase-field energy would require an argument that does not use uniform boundedness.
The observed power-law decay of the energy discrepancy with respect to $\lambda$ in Table 3 suggests a provable rate of the form $\|E_\lambda-E_0\|\sim\lambda^p$, which could guide the choice of the initial bracket in the bisection algorithm.
The constrained-optimization reformulation in Remark 2 could be solved directly instead of by bisection, potentially yielding even smaller weights and therefore energies closer to the original.
The weighting idea should transfer to exponential or logarithmic auxiliary variables and to other linear multi-step or Runge–Kutta discretizations, since the stability proofs are variational and the scalar equation remains one-dimensional.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The punchline: Theorem 3, the central existence guarantee, is false as stated. I checked the stress-test counterexample and it holds. Take L=G=1, τ=1, φ^n=4, E_N(φ)=cos φ, C=1.410. Then the scalar equation (3.7) is solvable at λ=1, but for λ=0.99 it has no real root, and by continuity no λ<1 sufficiently close to 1 works. E_N is bounded and smooth, so boundedness doesn't rescue the claim. The proof's inequality (3.12) treats f_min as a constant independent of λ, but f_min(λ) vanishes quadratically as λ→1 while the g-term is only O(1−λ), so the bound cannot hold. This is a load-bearing failure, not a corner case: the abstract and introduction specifically advertise solvability for any time step with a sufficiently large weight.

Credit where it's due. The weighting idea itself is a natural and useful interpolation between the two SAV families, and the discrete energy E=1/2(φ,Lφ)+λ(r^2−C)+(1−λ)E_N[φ] is well chosen. Theorem 4's dissipation identity is clean, no fitted constants. The second-order CN extension is straightforward. The numerics are thorough: temporal convergence tables, several 2D interface evolutions including the fractional Cahn–Hilliard case, and a 3D torus. The λ_min plots give concrete evidence that the plain Lagrange multiplier scheme loses solvability at practical time steps.

Soft spots. The adaptive bisection for λ_min is heuristic; Theorem 5 gives only conditional monotonicity, and the algorithm probes for solvability without a rigorous guarantee. The original-energy stability in Theorem 4 depends on an a posteriori inequality that the experiments never check. Boundedness of E_N is a genuine restriction; the tests enforce it by truncating the double-well at δ=5, but the counterexample shows boundedness alone is not enough for Theorem 3. Minor: Algorithm 2 references Eq. (3.7) instead of (3.32), and no code or data is posted.

Who this is for: numerical analysts working on SAV schemes and phase-field computations. Read it for the weighted-energy idea and the numerical comparisons; don't rely on Theorem 3. The paper deserves a serious referee because the flaw is subtle and the method may be repairable, but the current version should not be accepted without a corrected or weakened existence statement.

Referee Report

3 major / 4 minor

Summary. The paper proposes a weighted scalar auxiliary variable (SAV) method for gradient flows, unifying the nonlinear energy-based SAV approach (λ=1) and the Lagrange multiplier SAV approach (λ=0) through a weight coefficient λ∈[0,1]. Backward Euler and Crank–Nicolson time discretizations are formulated, leading to nonlinear scalar equations (3.7) and (3.32). The main theoretical claims are: (i) for any time step τ, solutions exist for all λ in some interval (λ_bar,1] (Theorem 3 and, without proof, Theorem 6); (ii) the modified energy (3.14) dissipates unconditionally, and the original energy dissipates under an additional condition (Theorems 4 and 7); and (iii) an adaptive bisection algorithm selects the smallest feasible λ. Numerical experiments on Allen–Cahn and Cahn–Hilliard type equations with a truncated quartic potential demonstrate first- and second-order accuracy, mass conservation, and smaller energy deviation for smaller λ.

Significance. The unification of the two SAV formulations is conceptually attractive, and the discrete energy-dissipation identities in Theorems 4 and 7 are derived cleanly. The numerical study is extensive and supports the practical observation that smaller λ reduces the deviation between the modified and original energies, and that the Lagrange multiplier limit can fail for large τ. However, the central existence theorem for arbitrary τ is false as stated; a concrete bounded nonlinearity gives no solution for λ close to 1 even though λ=1 is solvable. Since the claimed solvability guarantee is the main theoretical contribution and is also used to justify the adaptive algorithm, the paper in its current form does not establish its headline claim. With a corrected existence statement and a revised algorithm justification, the underlying weighting idea could still be a useful contribution.

major comments (3)

[Section 3.1, Theorem 3 and Eq. (3.12)] Theorem 3 is false as stated; the proof's sufficient condition (3.12) is invalid because f_min depends on λ. A scalar counterexample: take L=G=1, τ=1, φ^n=4, E_N(φ)=cos φ, and choose C = -sin 4 - cos 4 ≈ 1.4104 so that r^n = sqrt(cos 4 + C) and f_1(1)=2r^n + (H(φ^n), p^{n+1}-φ^n)/r^n = 0. With p^{n+1}=2 and q^{n+1} = -H(φ^n)/(2r^n) ≈ -0.4351, the equation (3.7) at λ=1 has only the root x=0, while at λ=0.99 the left-hand side h(x)=2.3586x^2+0.017x+0.01(cos(2-0.4351x)-cos 4) is strictly positive for all real x (global minimum ≈2.3×10^-3). Hence (3.7) has no solution for λ=0.99, and by continuity none for all λ<1 sufficiently close to 1, so no λ_bar<1 can have the property stated. The flaw is that f_min in (3.12) can vanish quadratically as λ→1 when f_1(1)=0, while the bound 2M(1-λ) is only linear. Theorem 6, whose proof is omitted, inherits this problem.
[Section 3.1, Theorem 3 and Section 4, Eq. (4.2)] The statement of Theorem 3 does not include the boundedness assumption |E_N|≤M that is introduced in its proof, and the paper's main numerical model (4.1) has E_N(φ)=∫(1/4)(φ^2-1-γ)^2 dx, which is unbounded on the natural function space. The truncated potential (4.2) makes E_N bounded, but the paper does not establish convergence of the truncated model to the original one as δ→∞ (nor as γ varies). Thus as written the theoretical solvability guarantee applies only to the truncated model, while the abstract presents it as a property of the weighted SAV method for gradient flows with bounded nonlinear energy. This gap should be stated explicitly in the theorem, and either the analysis extended or the numerical examples reinterpreted as solving the truncated problem.
[Section 3.1, Algorithm 1 and the paragraph after Theorem 5] The bisection search for λ_min assumes that solvability of (3.7) is monotone in λ in the sense that solvable values form an interval (λ_bar,1]. This is exactly the statement that Theorem 3 was supposed to provide, and it is false by the counterexample above. The assertion after Theorem 5 that 'there is no λ<λ_a that guarantees a solution' is not a consequence of Theorem 5(ii): that result only says that if a solution at λ* satisfies 2r*(r*-r^n)<E_N[φ*]-E_N[φ^n], then every λ∈[λ*,1] is solvable; it does not exclude isolated solvable values below λ*. Consequently, when the algorithm encounters an unsolvable midpoint and moves the bracket upward, it may discard all solvable λ, and the returned λ is not certified to be the minimal one. Algorithm 2 and the analogous claim for the CN scheme require the same correction.

minor comments (4)

[Throughout] There are several typographical errors, including 'utiliezd' in the Introduction, 'oganized' in the Introduction, and 'form' instead of 'from' in the step descriptions of Algorithms 1 and 2; these should be corrected.
[Section 3.2, Algorithm 2] In Algorithm 2, the solvability checks inside the bisection loop refer to 'Eq. (3.7)', but they should refer to the CN scalar equation (3.32).
[Section 3.2, Theorem 6] Theorem 6 is stated without a proof; given that the corresponding first-order theorem is false as stated, the sentence 'We will omit the detailed proof' is not acceptable unless the corrected first-order result is supplied and the analogous proof is shown to carry over.
[Section 4.3 and Eq. (3.14)] The statement that smaller λ gives a modified energy closer to the original is essentially a restatement of the definition (3.14) combined with the discrete relation between r and E_N; the numerical confirmation in Table 3 is therefore expected and should be presented as verification of the formula rather than as a new discovery.

Circularity Check

1 steps flagged · score 2.0 of 10

One mild self-definitional tautology in the energy-closeness claim; the core existence/stability derivations are self-contained, with the Theorem 3 gap being a correctness issue rather than circularity.

self definitional [Section 3.1, Theorem 4 and following paragraph, Eq. (3.14)]
"As shown in Eqs. (3.13)-(3.14) of the previous theorem, a smaller value of λ leads to a dissipation of discrete energy closer to E[φ n]."

By definition (3.14), E[φ^n,r^n] = 1/2(φ^n,Lφ^n) + λ((r^n)^2 - C) + (1-λ)E_N[φ^n]. Setting λ=0 makes the stabilized energy coincide with the original energy E[φ^n]=1/2(φ^n,Lφ^n)+E_N[φ^n], and λ=1 recovers the nonlinear-energy SAV energy. The stated closeness to the original energy as λ decreases is therefore read directly off the interpolation written into the definition. It does not follow from the scheme's dynamics or from any external benchmark, so presenting it as an achieved advantage is a definitional property rather than a derived prediction.

full rationale

The derivation chain is largely self-contained. The weighted scheme (3.3) and its reduced scalar equation (3.7) are stated explicitly; Theorems 3 and 4 follow, in the paper's intended argument, from the algebraic structure of (3.7) and the identity (3.18), using only the boundedness |E_N|≤M, positivity of (I+τGL), and the sign of (H(φ^n),q^{n+1}) from (3.11). No constant is fitted to benchmark data, no auxiliary variable is renamed as a prediction, and the only external existence result cited, [35], is used for the λ=0 small-τ baseline rather than as the proof of the new claim. The cited works by the same authors ([22,23,24,25,28,29]) are background, context, or numerical methodology, not load-bearing uniqueness or existence theorems. The one place with 'by construction' flavor is the assertion after Theorem 4 that smaller λ gives a discrete energy closer to E[φ^n]; since (3.14) defines that energy as a λ-interpolation, this is a mild definitional tautology. It does not contaminate the main existence/stability claims. Separately, the skeptical counterexample indicates Theorem 3's estimate (3.12) is not rigorous because f_min(λ) depends on λ and can vanish as λ→1; this is a mathematical correctness gap, not a circular reduction, and under the hard rules it is not counted in the circularity score.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method's central guarantees rest on the boundedness of E_N, the positivity of the SAV radicand, and the invertibility of I+tau G L. The adaptive weight lambda and the shift C are free parameters of the construction, and the truncation delta is a free parameter that reconciles the boundedness assumption with the unbounded quartic potential used in the experiments.

free parameters (3)

weight coefficient lambda_min = adaptive, per time step via bisection (Algorithm 1 and 2)
Interpolates between nonlinear SAV (lambda=1) and Lagrange multiplier SAV (lambda=0). The algorithm picks the smallest lambda that keeps the scalar equation solvable, so it is a free, adaptively chosen parameter of the method.
SAV shift constant C = not specified; chosen so E_N+C>0
Standard SAV parameter that defines r=sqrt(E_N+C). It appears in the modified energy (3.14), so it influences the discrete energy, but it is not fitted to data.
truncation parameter delta = 5 in numerical experiments
Introduced in Eq. (4.2) to make E_N bounded and satisfy the hypothesis of Theorem 3. It modifies the energy for |phi|>delta, so the simulated energy is the truncated energy, not the original quartic energy.

assumptions (4)

domain assumption The nonlinear energy E_N is uniformly bounded: there exists M such that |E_N|<=M
Stated and used in the proof of Theorem 3 to bound g(x;lambda) and derive the sufficient condition (3.12). The numerical experiments enforce it via the truncation (4.2), which changes the energy outside |phi|>delta.
standard math The operator I+tau G L is invertible for every tau>0
G is positive and L is non-negative, so I+tau G L is positive definite. Used to solve (3.6) uniquely.
domain assumption The radicand E_N[phi]+C is positive for all relevant phi
Required for the real scalar auxiliary variable r=sqrt(E_N+C) in (2.1). The constant C is chosen by hand to satisfy this.
domain assumption The truncation F_delta in (4.2) is C^2 and shares stationary points with the original potential for the parameter ranges used
Justifies that the numerical experiments with truncation approximate the intended phase-field model, and that the boundedness assumption holds in practice.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A weighted scalar auxiliary variable method for solving gradient flows: bridging the nonlinear energy-based and Lagrange multiplier approaches." pith.science (2026). https://pith.science/paper/GZD6VTLU

@misc{pith2026241117403,
  author       = {Pith},
  title        = {Pith review of: A weighted scalar auxiliary variable method for solving gradient flows: bridging the nonlinear energy-based and Lagrange multiplier approaches},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GZD6VTLU}},
  note         = {Machine review of arXiv:2411.17403}
}

read the original abstract

Two primary scalar auxiliary variable (SAV) approaches are widely applied for simulating gradient flow systems, i.e., the nonlinear energy-based approach and the Lagrange multiplier approach. The former guarantees unconditional energy stability through a modified energy formulation, whereas the latter preserves original energy stability but requires small time steps for numerical solutions. In this paper, we introduce a novel weighted SAV method which integrates these two approaches for the first time. Our method leverages the advantages of both approaches: (i) it ensures the existence of numerical solutions for any time step size with a sufficiently large weight coefficient; (ii) by using a weight coefficient smaller than one, it achieves a discrete energy closer to the original, potentially ensuring stability under mild conditions; and (iii) it maintains consistency in computational cost by utilizing the same time/spatial discretization formulas. We present several theorems and numerical experiments to validate the accuracy, energy stability and superiority of our proposed method.

Figures

Figures reproduced from arXiv: 2411.17403 by the authors.

**Figure 2.** (a): The evolution of weighting coefficient [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗

**Figure 3.** Several snapshots of curve I simulated using the weig [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: The evolution of (a) weighting coefficient [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]

**Figure 5.** Figure 5: Several snapshots of curve II simulated using the SAV [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

**Figure 6.** Figure 6: The evolution of (a) weighting coefficient [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]

**Figure 7.** Figure 7: The evolution of (a) normalized energy, (b) relative [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

**Figure 8.** Figure 8: Several snapshots of curve IV simulated using the wei [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

**Figure 9.** Figure 9: The evolution of (a) weighting coefficient [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]

**Figure 10.** Figure 10: Morphological evolution of an initial torus at six t [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]

**Figure 11.** Figure 11: The evolution of (a) weighting coefficient [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

[1]

Ainsworth, M., Mao, Z.: Analysis and approximation of a fr actional Cahn–Hilliard equation. SIAM J. Numer. Anal. 55, 1689–1718 (2017)

work page 2017
[2]

Akrivis, G., Li, B., Li, D.: Energy-decaying extrapolate d RK-SA V methods for the Allen–Cahn and Cahn–Hilliard equations. SIAM J. Sci. Comput. 41, A3703–A3727 (2019) 19

work page 2019
[3]

Antoine, X., Shen, J., Tang, Q.: Scalar auxiliary variabl e/Lagrange multiplier based pseudospectral schemes for th e dynamics of nonlinear Schr¨ odinger/Gross–Pitaevskii equations. J. Comput. Phys. 437, 110328 (2021)

work page 2021
[4]

Backofen, R., Wise, S., Salvalaglio, M., Voigt, A.: Conve xity splitting in a phase ﬁeld model for surface diﬀusion. In t. J. Numer. Anal. Mod. 16, 192–209 (2019)

work page 2019
[5]

Bai, X., Sun, J., Shen, J., Yao, W., Guo, Z.: A Ginzburg–Lan dau–H − 1 model and its SA V algorithm for image inpainting. J. Sci. Comput. 96, 40 (2023)

work page 2023
[6]

Bao, W., Jiang, W., Li, Y.: A symmetrized parametric ﬁnite element method for anisotropic surface diﬀusion of closed curves. SIAM J. Numer. Anal. 61, 617–641 (2023)

work page 2023
[7]

Bao, W., Jiang, W., W ang, Y., Zhao, Q.: A parametric ﬁnite e lement method for solid-state dewetting problems with anisotropic surface energies. J. Comput. Phys. 330, 380–400 (2017)

work page 2017
[8]

Chen, H., Mao, J., Shen, J.: Optimal error estimates for th e scalar auxiliary variable ﬁnite-element schemes for grad ient ﬂows. Numer. Math. 145, 167–196 (2020)

work page 2020

Show all 42 references

[9]

Chen, C., Yang, X.: Fast, provably unconditionally energ y stable, and second-order accurate algorithms for the anisotropic Cahn–Hilliard model. Comput. Methods Appl. Me ch. Engrg. 351, 35–59 (2019)

2019
[10]

Cheng, Q., Liu, C., Shen, J.: A new Lagrange multiplier ap proach for gradient ﬂows. Comput. Methods Appl. Mech. Engrg. 367, 113070 (2020)

2020
[11]

Cheng, Q., Liu, C., Shen, J.: Generalized SA V approaches for gradient systems. J. Comput. Appl. Math. 394, 113532 (2021)

2021
[12]

Cox, S., Matthews, P.: Exponential time diﬀerencing for stiﬀ systems. J. Comput. Phys. 176, 430–455 (2002)

2002
[13]

Handbook of Numerical Analysis 21, 425–508 (2020)

Du, Q., Feng, X.: The phase ﬁeld method for geometric movi ng interfaces and their numerical approximations. Handbook of Numerical Analysis 21, 425–508 (2020)

2020
[14]

Du, Q., Liu, C., W ang, X.: Simulating the deformation of v esicle membranes underelastic bending energy in three dimensions. J. Comput. Phys. 212, 757–777 (2006)

2006
[15]

A.: Numerical analysis of a contin uum model of phase transition

Du, Q., Nicolaides, R. A.: Numerical analysis of a contin uum model of phase transition. SIAM J. Numer. Anal. 28, 1310–1322 (1991)

1991
[16]

Eyre, D.: Unconditionally gradient stable time marchin g the Cahn–Hilliard equation. Mater. Res. Soc. Sympos. Proc . 529, 39–46 (1998)

1998
[17]

Feng, X., Li, B., Ma, S.: High-order mass- and energy-con serving SA V-Gauss collocation ﬁnite element methods for the nonlinear Schr¨ odinger equation. SIAM J. Numer. Anal. 59, 1566–1591 (2021)

2021
[18]

Fu, Z., Yang, J.: Energy-decreasing exponential time di ﬀerencing Runge–Kutta methods for phase-ﬁeld models. J. Comput. Phys. 454, 110943 (2022)

2022
[19]

Gong, Y., Hong, Q., W ang, Q.: Supplementary variable met hod for thermodynamically consistent partial diﬀerential equations. Comput. Methods Appl. Mech. Engrg. 381, 113746 (2021)

2021
[20]

Gong, Y., Zhao, J., W ang, Q.: Arbitrarily high-order unc onditionally energy stable SA V schemes for gradient ﬂow models. Comput. Phys. Commun. 249, 107033 (2020)

2020
[21]

Hong, Q., W ang, Q., Gong, Y.: High-order supplementary v ariable methods for thermodynamically consistent partial diﬀerential equations. Comput. Methods Appl. Mech. Engrg. 416, 116306 (2023)

2023
[22]

Z.: An unconditionally energy stable scheme for simulating wrinkling phenomena of elastic thin ﬁlms on a compliant substrate

Huang, Q.-A., Jiang, W., Yang, J. Z.: An unconditionally energy stable scheme for simulating wrinkling phenomena of elastic thin ﬁlms on a compliant substrate. J. Comput. Phy s. 388, 123–143 (2019)

2019
[23]

Z.: An eﬃcient and uncon ditionally energy stable scheme for simulating solid-stat e dewetting of thin ﬁlms with isotropic surface energy

Huang, Q.-A., Jiang, W., Yang, J. Z.: An eﬃcient and uncon ditionally energy stable scheme for simulating solid-stat e dewetting of thin ﬁlms with isotropic surface energy. Commu n. Comput. Phys. 26, 1444–1470 (2019)

2019
[24]

Z., Yuan, C.: A structur e-preserving, upwind-SA V scheme for the degenerate Cahn–Hilliard equation with applications to simulating su rface diﬀusion

Huang, Q.-A., Jiang, W., Yang, J. Z., Yuan, C.: A structur e-preserving, upwind-SA V scheme for the degenerate Cahn–Hilliard equation with applications to simulating su rface diﬀusion. J. Sci. Comput. 97, 64 (2023)

2023
[25]

Z., Zhang, G.: Linear mu lti-step methods and their numerical stability for solving gradient ﬂow equations

Huang, Q.-A., Jiang, W., Yang, J. Z., Zhang, G.: Linear mu lti-step methods and their numerical stability for solving gradient ﬂow equations. Adv. Comput. Math. 49, 39 (2023)

2023
[26]

Huang, F., Shen, J., Yang, Z.: A highly eﬃcient and accura te new scalar auxiliary variable approach for gradient ﬂows. SIAM J. Sci. Comput. 42, A2514–A2536 (2020)

2020
[27]

Huang, J., Yang, C., W ei, Y.: Parallel energy-stable sol ver for a coupled Allen–Cahn and Cahn–Hilliard system. SIAM J. Sci. Comput. 42, C294–C312 (2020)

2020
[28]

Huang, Q.-A., Yuan, C., Zhang, G., Zhang, L.: A computati onally optimal relaxed scalar auxiliary variable approach for solving gradient ﬂow systems. Comput. Math. Appl. 156, 64–73 (2024)

2024
[29]

Huang, Q.-A., Zhang, G., W u, B.: Fully-discrete energy- preserving scheme for the space-fractional Klein–Gordon equation via Lagrange multiplier type scalar auxiliary var iable approach. Math. Comput. Simulat. 192, 265–277 (2022)

2022
[30]

Jiang, W., Su, C., Zhang, G.: A convexity-preserving and perimeter-decreasing parametric ﬁnite element method for the area-preserving curve shortening ﬂow. SIAM J. Numer. An al. 61, 1989–2010 (2023)

2023
[31]

Jiang, W., Su, C., Zhang, G.: A second-order in time, BGN- based parametric ﬁnite element method for geometric ﬂows of curves. J. Comput. Phys. 514, 113220 (2024)

2024
[32]

Jiang, M., Zhang, Z., Zhao, J.: Improving the accuracy an d consistency of the scalar auxiliary variable (SA V) method with relaxation. J. Comput. Phys. 456, 110954 (2022)

2022
[33]

Liu, Z., Li, X.: The exponential scalar auxiliary variab le (E-SA V) approach for phase ﬁeld models and its explicit computing. SIAM J. Sci. Comput. 42, B630–B655 (2020)

2020
[34]

CSIAM Trans

Liu, Z., Li, X.: A novel Lagrange multiplier approach wit h relaxation for gradient ﬂows. CSIAM Trans. Appl. Math. 5, 110–141 (2024)

2024
[35]

Onuma, K., Sato, S.: Existence results on Lagrange multi plier approach for gradient ﬂows and application to opti- mization. Japan J. Ind. Appl. Math. 41, 165–189 (2024)

2024
[36]

Pei, L., Li, Y.: A structure-preserving parametric ﬁnit e element method for area-conserved generalized curvature ﬂow. J. Sci. Comput. 96, 6 (2023)

2023
[37]

Qian, Y., Yang, Z., W ang, F., Dong, S.: gPA V-based uncond itionally energy-stable schemes for the Cahn–Hilliard equation: Stability and error analysis. Comput. Methods Ap pl. Mech. Engrg. 372, 113444 (2020)

2020
[38]

Shen, J., Xu, J., Yang, J.: The scalar auxiliary variable (SA V) approach for gradient ﬂows. J. Comput. Phys. 353, 407–416 (2018) 20

2018
[39]

SIAM Rev

Shen, J., Xu, J., Yang, J.: A new class of eﬃcient and robus t energy stable scheme for gradient ﬂows. SIAM Rev. 61, 474–506 (2019)

2019
[40]

Tang, T., Yu, H., Zhou, T.: On energy dissipation theory a nd numerical stability for time-fractional phase-ﬁeld equations. SIAM J. Sci. Comput. 41, A3757–A3778 (2019)

2019
[41]

Yang, Z., Dong, S.: A roadmap for discretely energy-stab le schemes for dissipative systems based on a generalized auxiliary variable with guaranteed positivity. J. Comput. Phys. 404, 109121 (2020)

2020
[42]

East Asian J

Yoon, S., W ang, J., Lee, C., Yang, J., Park, J., Kim, H., Ki m, J.: Numerical investigation to the eﬀect of initial guess for phase-ﬁeld models. East Asian J. Appl. Math. 11, 618–646 (2021) 21

2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Ainsworth, M., Mao, Z.: Analysis and approximation of a fr actional Cahn–Hilliard equation. SIAM J. Numer. Anal. 55, 1689–1718 (2017)

work page 2017

[2] [2]

Akrivis, G., Li, B., Li, D.: Energy-decaying extrapolate d RK-SA V methods for the Allen–Cahn and Cahn–Hilliard equations. SIAM J. Sci. Comput. 41, A3703–A3727 (2019) 19

work page 2019

[3] [3]

Antoine, X., Shen, J., Tang, Q.: Scalar auxiliary variabl e/Lagrange multiplier based pseudospectral schemes for th e dynamics of nonlinear Schr¨ odinger/Gross–Pitaevskii equations. J. Comput. Phys. 437, 110328 (2021)

work page 2021

[4] [4]

Backofen, R., Wise, S., Salvalaglio, M., Voigt, A.: Conve xity splitting in a phase ﬁeld model for surface diﬀusion. In t. J. Numer. Anal. Mod. 16, 192–209 (2019)

work page 2019

[5] [5]

Bai, X., Sun, J., Shen, J., Yao, W., Guo, Z.: A Ginzburg–Lan dau–H − 1 model and its SA V algorithm for image inpainting. J. Sci. Comput. 96, 40 (2023)

work page 2023

[6] [6]

Bao, W., Jiang, W., Li, Y.: A symmetrized parametric ﬁnite element method for anisotropic surface diﬀusion of closed curves. SIAM J. Numer. Anal. 61, 617–641 (2023)

work page 2023

[7] [7]

Bao, W., Jiang, W., W ang, Y., Zhao, Q.: A parametric ﬁnite e lement method for solid-state dewetting problems with anisotropic surface energies. J. Comput. Phys. 330, 380–400 (2017)

work page 2017

[8] [8]

Chen, H., Mao, J., Shen, J.: Optimal error estimates for th e scalar auxiliary variable ﬁnite-element schemes for grad ient ﬂows. Numer. Math. 145, 167–196 (2020)

work page 2020