{"id":"0bafc565-3b99-478b-9036-c71a8d348660","arxiv_id":"1908.04910","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A finite element scheme for Cahn-Hilliard equations with dynamic boundary conditions is unconditionally energy stable, mass conservative, and proven to converge to weak solutions.","lead":"This paper builds a finite element method for the Cahn-Hilliard phase-field model with dynamic boundary conditions, eliminating the chemical potential variables to make each time step cheap. It proves the method conserves mass, dissipates energy, and converges to weak solutions of the continuous model.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Convergence is conditional on refinement condition (C); the adaptive regime in §5 is not covered by Theorem 4.4.","rationale":"I read the proof of Theorem 4.4 carefully, including the algebraic reduction from (2.9) to (2.23), the energy-stability argument in Lemma 3.3, the existence proof in Lemma 3.4, and the compactness and limit passage in Section 4. I found no mathematical error that invalidates the central claim as stated. The scheme is derived directly from the weak formulation, the discrete energy inequality is unconditional, and the convergence proof is standard for this class of convex-splitting finite element methods. The identified load-bearing issue is the refinement condition (C) in Lemma 4.2: without it, the interpolation-error terms in the dual-time-derivative bounds do not vanish, and the Aubin–Lions compactness argument cannot be completed. This condition is explicitly stated, so the theorem is internally consistent, but it limits the practical scope of the convergence result, especially because the simulations in Section 5 use adaptive time steps with very small τ on fixed meshes. The reader's weakest_assumption named exactly this condition, and I agree. Since the theorem is conditional on (C) and the paper does not claim convergence outside those assumptions, the appropriate verdict remains ACCEPT rather than REJECT or CONDITIONAL. Minor presentation issues such as the definition of Xκ in (1.16) appearing to require κ>1 rather than κ>0 and a missing absolute value in the set Mη in the proof of Corollary 4.1 are likely typos and do not affect the central argument.","tokens_in":31768,"tokens_out":31082,"duration_ms":325596,"concrete_test":"Run the §5.2 convergence study on a fixed quasi-uniform mesh with κ=1, taking time-step sequences that both satisfy and violate (C), e.g. τ ∝ h^4 and τ ∝ h^5. For each run, compute the left side of (4.8), specifically τ Σ ||∂τ φ||^2_{(H1)'}, or at least the interpolation-error term h^4/τ Σ ||∇(φ^n−φ^{n−1})||^2. If this quantity remains bounded as τ/h^4 → ∞, then condition (C) is not necessary and the theorem's scope could be widened; if it grows without bound, (C) is essential and the adaptive simulations are genuinely outside the proven convergence statement.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 4.4 relies on Lemma 4.2 for uniform bounds on the discrete time derivative ∂t φ in L2((H1)') and L2((H1(Γ))'). Equations (4.8) and (4.9) contain interpolation-error terms h^4/τ Σ ||∇(φ^n−φ^{n−1})||^2_{L2(Ω)} (κ>0) and h^2/τ Σ ||φ^n−φ^{n−1}||^2_{L2(Γ)} (κ=0), which are controlled only under condition (C). If (C) fails, the dual-space bound on ∂t φ is lost, so the Aubin–Lions compactness argument in Lemma 4.3 no longer applies and the passage to the limit in (4.12) is unjustified. This is an explicit assumption rather than a hidden error, but it is the least secure point of the convergence claim. The numerical section uses adaptive time stepping with τ as small as 6.3·10^−7 on a fixed mesh, so h^4/τ is not guaranteed to tend to zero; hence the practical adaptive regime is not covered by the proven theorem.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes and analyzes a finite element scheme for the Cahn–Hilliard equation with dynamic boundary conditions of Liu–Wu type, where an additional Cahn–Hilliard equation on the boundary is coupled to the bulk equation through a normal-derivative term. The main contributions are: (i) an efficient reformulation of the fully discrete scheme that eliminates the two chemical potentials and avoids the degeneracy of the monolithic system as the time step tends to zero; (ii) a discrete energy-stability result (Lemma 3.3) and an existence result for discrete solutions (Lemma 3.4); (iii) a convergence theorem (Theorem 4.4) showing that discrete solutions converge along a sequence (h,τ)→0, under a mesh-time coupling condition, to a weak solution of the continuous model, thereby also providing an alternative existence proof; and (iv) numerical experiments documenting mass conservation, energy dissipation, conditioning behavior, and experimental orders of convergence.","tokens_in":31899,"tokens_out":13047,"duration_ms":132122,"significance":"If the results are correct, the paper gives a valuable toolbox for a nontrivial boundary-coupled Cahn–Hilliard system: an implementable, unconditionally energy-stable scheme whose convergence is proven, and a pathway to weak solutions that complements the gradient-flow existence theory of Garcke and Knopf. The paper is honest about its assumptions and provides detailed proofs for the main estimates, including the delicate interpolation-error control in Lemma 4.2. The simulations are reproducible in principle and illustrate the practical conditioning advantages of the proposed formulation. The main limitations are the restrictive mesh-time condition (C) in the convergence theorem and a few proof gaps that need to be closed before the claims are fully rigorous.","major_comments":[{"comment":"The proof of Lemma 3.4 begins with \"this allows us to assume w.l.o.g. that 1^T M_Ω Φ^n = 1^T M_Ω Φ^{n-1} = 0,\" but mass conservation alone does not justify a reduction to zero mean because the discrete scheme (2.23) is not invariant under adding a constant to Φ. The later inequality (3.13) uses the equivalence of sqrt(Φ^T L_Ω Φ) with the discrete L2 norm, which holds only on the mean-zero subspace. To make the argument rigorous for arbitrary initial data, the proof should work in the affine space of fixed total mass m_0, using a Poincaré-type inequality with the fixed mean, or should show that the energy estimate bounds the mean; otherwise the existence claim for general initial data is not established. This is a load-bearing step for the central existence result, although it is fixable.","section":"§3, Lemma 3.4"},{"comment":"The statement of Lemma 3.4 lists only assumptions (T), (S1), (S2), and (P1), but the proof uses additional structure from (P2): the concavity inequality for G'_- with β, the polynomial-growth control of F'_-, and the discrete energy estimate of Lemma 3.3 that itself requires (P2). Either (P2) must be added to the lemma's hypotheses, or the proof must be modified to work under (P1) alone. As written, the statement and proof do not match.","section":"§3, Lemma 3.4 statement"},{"comment":"Theorem 4.4 is conditional on the mesh-time coupling (C): h^4/τ→0 when κ>0 and h^2/τ→0 when κ=0. This is an explicit assumption, but it is strong: it requires τ to be much larger than h^4 (respectively h^2), so arbitrarily small time steps are admissible only on correspondingly fine meshes. The adaptive simulations in Section 5.1 use time increments down to 6.3·10^-7 on fixed or adaptively refined meshes, and the paper does not discuss how the adaptive time-stepping relates to condition (C). The authors should clarify that the convergence theorem applies to simultaneous refinement paths satisfying (C) and that the adaptive simulations in Section 5 are numerical demonstrations rather than instances covered by Theorem 4.4.","section":"§4, Assumption (C) and §5.1"},{"comment":"The passage to the limit in the nonlinear potential terms for d=3 is only sketched: the text says \"The uniform bounds of φ_h^{τ,±} in L∞(0,T;H1(Ω)) provide enough regularity, to adapt the previously presented arguments to three spatial dimensions.\" For κ=0, the strong convergence (4.15f) on Γ holds only in L^s for s<4, while the cubic estimates (4.21)–(4.23) appear to require control in L^4(Γ). The exponent bookkeeping in three dimensions should be written out (or a precise embedding/interpolation argument given) so that the claimed validity of Theorem 4.4 for d=3 is fully supported.","section":"§4, Theorem 4.4, three-dimensional case"}],"minor_comments":[{"comment":"The definition of X_κ states \"κ>1\" in the first case and \"κ=0\" in the second, but the analysis throughout the paper treats all κ>0 as the surface-diffusion case. This appears to be a typo; it should read \"κ>0\".","section":"§1, Eq. (1.16)"},{"comment":"There are several typographical errors, including \"asssume\" in Section 2, \"satifies\" in Lemma 3.3, and the missing space in \"tuple(φ,µ,µ Γ)\" in Theorem 4.4. These should be corrected in a final revision.","section":"Throughout"},{"comment":"The experimental order of convergence on Γ is 1.1, notably lower than the order 2.3 measured on Ω. The authors should comment on this discrepancy, since it may indicate a boundary-related suboptimality in the scheme or in the error measurement.","section":"§5.2, Table 3"},{"comment":"Remark 4.5 claims that the convergence results carry over to Allen–Cahn-type boundary conditions, but it only lists the relevant uniform bounds and states that \"we are able to identify converging subsequences and pass to the limit.\" Since this is an additional model with a different boundary equation, a brief sketch of the convergence argument (or a reference to the analogous steps) would be helpful.","section":"§4, Remark 4.5"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and the central idea is potentially valuable. My main concern is the gap in Lemma 3.4: the zero-mean reduction is not justified as written, and since existence of discrete solutions is a headline result, this needs to be fixed before acceptance. The missing (P2) in the lemma statement is a smaller but related issue. The restriction imposed by condition (C) is stated honestly, but the relation between the adaptive numerics and the convergence theorem should be clarified. I do not see evidence of circularity or of hidden assumptions; the issues are local and likely repairable in a moderately extended revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth a serious look. What is actually new is the boundary Schur complement elimination that turns the coupled bulk/boundary Cahn-Hilliard system of Liu-Wu into a scheme that only needs a small positive definite linear solve on the boundary at each nonlinear iteration. The author proves unconditional energy stability, existence of discrete solutions for arbitrary time steps, and convergence of discrete solutions to weak solutions of the continuum model. The convergence proof is detailed and doubles as an alternate existence proof for the continuous problem, as stated.\n\nCredit where due: the algebraic reduction is clean and the condition-number comparison against the monolithic scheme makes a practical point, not a decorative one. The discrete energy estimate is straightforward but correct. The convergence argument follows standard Aubin-Lions machinery, and the author is upfront about the non-singular potential restriction and the refinement condition (C).\n\nThe soft spots are in proportion. The stress-test note is right about (C): Theorem 4.4 requires h^4/tau -> 0 (or h^2/tau when kappa=0), and the numerical section uses adaptive time stepping with tau as small as 6.3e-7 on a fixed mesh, so the proven theorem does not cover the practical adaptive regime. This is an explicit assumption rather than a hidden error, but it should be brought out more in the discussion. A second limitation is that the theory proves convergence but no rates; the EOC numbers in Section 5.2 are empirical. The w.l.o.g. mean-zero reduction in Lemma 3.4 deserves a sentence or two more. None of these break the central argument. The citation pattern is honest: prior work by Trautwein and Garcke-Knopf is cited and distinguished, and the claimed novelty holds.\n\nThe paper delivers a rigorous, practical scheme for a model that previously only had a monolithic discretization in a bachelor thesis. It is the kind of contribution that belongs in a numerical analysis journal. A serious referee should engage with it; with reasonable revisions, accept.","headline":"Solid numerical analysis paper that deserves peer review; the convergence theorem's refinement condition is real but explicit, and the adaptive numerics sit outside its stated scope.","tokens_in":32485,"tokens_out":2569,"would_cite":true,"duration_ms":25815,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q35","35G31","65M60","65M12"],"pacs":[],"model":"deepseek-v4-flash","headline":"A new finite element scheme for the Cahn–Hilliard equation with dynamic boundary conditions is proven to be unconditionally energy stable and convergent to weak solutions.","keywords":["Cahn-Hilliard equation","dynamic boundary conditions","finite element method","convergence","unconditional energy stability","convex-concave splitting","weak solutions"],"falsifier":"For a manufactured smooth solution of (1.1) on a convex polygon with a polynomial double-well potential, take mesh and time-step pairs with $h^4/\\tau\\to 0$ and measure the discrete $L^2$ error; if the error does not tend to zero along such a sequence, the convergence claim of Theorem 4.4 is false.","tokens_in":31488,"feed_emoji":"🧮","tokens_out":11855,"duration_ms":108540,"temperature":0.7,"pith_summary":"The paper proposes a finite element discretization for a Cahn–Hilliard phase-separation model in which the boundary has its own Cahn–Hilliard dynamics, coupled to the interior through the normal derivative of the phase field. The central claim is that an algebraic elimination of the chemical potentials yields a scheme that is unconditionally energy stable, admits discrete solutions for arbitrary time steps, and converges to weak solutions of the continuous model as both mesh size and time step go to zero. The convergence theorem also supplies an alternative route to proving existence of weak solutions. Numerical experiments illustrate mass conservation, energy dissipation, and experimental convergence orders of about two in space and 1.6 in time.","feed_headline":"Cahn-Hilliard scheme proven convergent with dynamic boundaries","feed_subtitle":"The method conserves mass, dissipates energy, and yields weak solutions of the continuous model.","key_machinery":"The load-bearing object is an explicit algebraic reduction of the coupled interior–boundary chemical-potential system. Starting from the monolithic system (2.15), the paper separates boundary and interior degrees of freedom, eliminates the interior chemical potential and the boundary chemical potential, and obtains a small symmetric positive definite boundary system with matrix $m L_\\Omega|_{\\Gamma\\times\\Gamma}+m_\\Gamma M_\\Omega|_{\\Gamma\\times\\Gamma}M_\\Gamma^{-1}L_\\Gamma M_\\Gamma^{-1}M_\\Omega|_{\\Gamma\\times\\Gamma}$. The resulting single nonlinear equation (2.23) for the new phase-field values has a structure that does not degenerate as $\\tau\\to0$, while still satisfying the compatibility constraint that makes it equivalent to the original coupled system. This reduction turns the numerical analysis into that of a standard Cahn–Hilliard scheme: a convex–concave splitting in time supplies unconditional energy stability, interpolation estimates supply the missing regularity, and a compactness argument identifies the limit.","core_discovery":"The paper's main result, Theorem 4.4, states that under assumptions on the mesh, the potentials, and the initial data, discrete solutions of scheme (2.9)/(2.23) converge along sequences $(h,\\tau)\\to 0$ to a triple $(\\varphi,\\mu,\\mu_\\Gamma)$ that solves the weak formulation (4.18) of the model. The potentials $F$ and $G$ must admit a convex–concave splitting with polynomial part of degree four plus a globally Lipschitz part; the initial data must have bounded energy; and the mesh and time step must satisfy $h^4/\\tau\\to0$ when the surface-diffusion coefficient $\\kappa>0$, or $h^2/\\tau\\to0$ when $\\kappa=0$. In the same framework the scheme conserves the phase-field mass in the domain and on the boundary separately, dissipates the discrete total energy, and has at least one solution for every time step. Passing to the limit in the scheme therefore also constitutes an existence proof for weak solutions.","pith_inferences":["The refinement condition $h^4/\\tau\\to0$ (or $h^2/\\tau\\to0$ when $\\kappa=0$) is a strong mesh–time coupling that adaptive time stepping may violate; testing convergence on fixed meshes as $\\tau\\to0$ would reveal whether this condition is an artifact of the proof.","The boundary-only auxiliary system is symmetric positive definite, so in three dimensions the scheme's cost is dominated by a boundary problem; this suggests dedicated multilevel or preconditioned iterative solvers could make the method competitive for large-scale simulations.","The same elimination strategy might extend to related bulk–surface phase-field models, such as models with non-instantaneous adsorption or reaction-rate transfer between boundary and bulk, wherever a compatibility relation between chemical potentials can be derived.","The assumptions exclude singular potentials like logarithmic or double-obstacle wells; experimenting with penalized but smooth approximations may indicate whether the convergence theory can be pushed to singular potentials."],"forward_implications":["Discrete solutions exist for any time step size, without a CFL-type restriction, because the existence argument uses only the a priori energy estimate.","The discrete total energy (bulk plus surface) decreases along solutions, and the mean phase-field values in the domain and on the boundary are conserved separately.","Along mesh and time-step sequences satisfying condition (C), the discrete solutions yield weak solutions of the continuous model, so the scheme provides a constructive existence proof for the boundary-dynamics model.","The same algebraic reduction and convergence analysis carry over to Allen–Cahn-type dynamic boundary conditions, including a dynamic contact-angle boundary condition.","In the experiments, the phase field converges with experimental order about 2.3 with respect to mesh size in the bulk and 1.1 on the boundary, and order about 1.6 with respect to the time step."],"supporting_citations":[{"why":"Supplies the continuous bulk–surface Cahn–Hilliard model that the scheme discretizes and analyzes.","marker":"[38]"},{"why":"Provides the existence framework for weak solutions and the regularity strategy that Corollary 4.1 follows.","marker":"[26]"},{"why":"Provides the interpolation, inverse-estimate, and approximation properties of the piecewise-linear finite element spaces used throughout the convergence proof.","marker":"[10]"},{"why":"Supplies the functional-analytic inequality used to upgrade gradient control to full $H^1$ control of the chemical potentials.","marker":"[2]"},{"why":"Supplies the stability of the $L^2$ projection in $H^1$, needed for the dual-space bound on the discrete time derivative.","marker":"[9]"},{"why":"Supports the convex–concave splitting treatment of the double-well potential that yields unconditional energy stability.","marker":"[31]"},{"why":"Defines the dynamic contact-angle boundary condition that the extended scheme in Remark 4.5 covers.","marker":"[47]"}],"fun_headline_variants":["Convergent FE for Cahn-Hilliard with dynamic BCs","Energy-stable convergent FE for Cahn-Hilliard dynamic BC","Cahn-Hilliard scheme with dynamic BC: proved convergent","Dynamic-boundary Cahn-Hilliard: convergent FE scheme","Finite element scheme for dynamic-boundary Cahn-Hilliard proven convergent"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The convergence proof requires the mesh size $h$ and time step $\\tau$ to satisfy $h^4/\\tau\\to0$ when $\\kappa>0$ and $h^2/\\tau\\to0$ when $\\kappa=0$; if a computation drives $\\tau$ to zero on a fixed mesh, the bound on the discrete time derivative in the dual space is lost and the compactness argument that produces the weak solution no longer applies.","fun_headline_variants_meta":{"raw":{"variants":["Convergent FE for Cahn-Hilliard with dynamic BCs","Energy-stable convergent FE for Cahn-Hilliard dynamic BC","Cahn-Hilliard scheme with dynamic BC: proved convergent","Dynamic-boundary Cahn-Hilliard: convergent FE scheme","Finite element scheme for dynamic-boundary Cahn-Hilliard proven convergent"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000718,"raw_usage":{"total_tokens":3220,"prompt_tokens":938,"completion_tokens":2282,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":2204}},"tokens_in":554,"tokens_out":2282,"duration_ms":16673,"temperature":1.0,"reasoning_tokens":2204,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:29:23.982144+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a manufactured smooth solution of (1.1) on a convex polygon with a polynomial double-well potential, take mesh and time-step pairs with $h^4/\\tau\\to 0$ and measure the discrete $L^2$ error; if the error does not tend to zero along such a sequence, the convergence claim of Theorem 4.4 is false.","supporting_citations":[{"cited_title":"Liu and H","cited_arxiv_id":null,"evidence_quote":"Supplies the continuous bulk–surface Cahn–Hilliard model that the scheme discretizes and analyzes."},{"cited_title":"Garcke and P","cited_arxiv_id":null,"evidence_quote":"Provides the existence framework for weak solutions and the regularity strategy that Corollary 4.1 follows."},{"cited_title":"Brenner and L","cited_arxiv_id":null,"evidence_quote":"Provides the interpolation, inverse-estimate, and approximation properties of the piecewise-linear finite element spaces used throughout the convergence proof."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the functional-analytic inequality used to upgrade gradient control to full $H^1$ control of the chemical potentials."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the stability of the $L^2$ projection in $H^1$, needed for the dual-space bound on the discrete time derivative."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports the convex–concave splitting treatment of the double-well potential that yields unconditional energy stability."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the dynamic contact-angle boundary condition that the extended scheme in Remark 4.5 covers."}],"review_version":1}