{"id":"cff172b6-f47d-4da5-86ee-03a9e5410f4b","arxiv_id":"2607.26341","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A fixed-grid correction method with an interface density solves moving-domain Robin advection-diffusion, backed by a 1D convergence proof and 2D numerical tests.","lead":"This paper presents a numerical method for advection-diffusion equations on moving domains with Robin boundary conditions, using a fixed Cartesian grid and an interface correction rather than remeshing. It proves first-order-in-time and second-order-in-space convergence for the one-dimensional scheme and demonstrates accuracy and mesh-independent solver cost in two dimensions.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 1D convergence proof (Theorem 4.13) has a scaling gap: Lemma 4.12's correction is bounded only as O(h), not O(h^2), because Eq. (4.104) contains an unabsorbed 1/h; the claimed O(tau+h^2) error bound is not established.","rationale":"Good-faith reading: the paper is a serious numerical analysis contribution with a credible scheme and a genuinely ambitious 1D analysis. The 2D collocation concern that the reader identified is real but secondary, because Section 4 explicitly limits the proof to one dimension and Section 5 treats 2D as numerical evidence. The strongest claim is the proved 1D estimate in Theorem 4.13. On close reading, Lemma 4.12's key construction has a factor-of-h mismatch: the source in (4.104) is O(tau h), so Theorem 4.7 yields O(h), not O(h^2). If this reading is correct, the paper's only convergence proof is incomplete. The numerical results may still be right, but the central claim 'first-order convergence in time and second-order convergence in space are proved' is unsupported. A single analytical/computational check can settle the issue. Because the proof gap undermines the main analytical contribution, I would move the reader's CONDITIONAL verdict to REJECT for the current version, with the caveat that a repaired Lemma 4.12 or an explicit derivation of the missing cancellation would change the verdict.","tokens_in":1239,"tokens_out":1970,"duration_ms":164736,"concrete_test":"Analytical check: substitute (4.104) and the definition of hat g^n into Theorem 4.7 verbatim. The maximum-principle source is E^n hat g^n/(h Q^n) = O(h), so Lemma 4.12's O(h^2) bound is false unless a hidden cancellation is identified. Computational check: for the Section 5.1 manufactured solution, compute r_{2,j(n)}^n and r_4^n from Lemma 4.9, solve (4.104) for hat b_c^n, and measure sup_n ||R^n hat b_c^n||_inf. Linear decay (order ~1) confirms the gap; second-order decay (order ~2) would disprove the concern and would require identifying the missing factor explicitly.","verdict_should_be":"REJECT","load_bearing_attack":"Lemma 4.12 is the load-bearing step of the paper's only proof. It defines hat b_c^n by Eq. (4.104): R^n K^n (R^n)^T R^n hat b_c^n - R^n M^{n-1}(R^{n-1})^T R^{n-1} hat b_c^{n-1} = tau/(h Q^n) R^n E^n hat g^n, with hat g^n = (h Q^n r_{2,j(n)}^n / E_{j(n)}^n) + r_4^n = O(h^2). Since R^n E^n / Q^n is uniformly bounded, the source has size O(h), not O(h^2), and Theorem 4.7 gives ||R^n hat b_c^n||_inf <= C e^{CT} sup |hat g^n| incorrectly; the factor 1/h is dropped. Also, the sign of hat g^n as printed reproduces, rather than cancels, the O(tau h) term in Corollary 4.10. Therefore the O(h) boundary-localized consistency error is not removed, and Theorem 4.13's bound does not follow. This affects the central claim 'first-order in time and second-order in space are proved' more directly than the absence of a 2D proof.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a Cartesian grid (embedded boundary) method for advection-diffusion equations with Robin boundary conditions on moving domains. The moving-domain problem is reformulated as an interface problem on a fixed box, with an unknown single-layer density introduced on the moving interface. A local polynomial correction function is built in a narrow band via 6x6 collocation problems at each interface point; the correction enters only the right-hand side, preserving the standard bulk finite-difference operator. The coupled system is solved by eliminating correction and bulk unknowns and applying matrix-free GMRES to the interface-density Schur complement. The main theoretical result is Theorem 4.13: a fully discrete l-infinity error estimate O(tau + h^2) for a one-dimensional version of the scheme, under the mesh constraint 1/2 h^2 <= tau < h / max|gamma_dot| and monotone interface motion. Numerical experiments in 1D confirm the predicted rates; 2D tests on circular and star-shaped interfaces and an active-transport problem show near-second-order self-convergence and mesh-insensitive GMRES counts.","tokens_in":28251,"tokens_out":23965,"duration_ms":220992,"significance":"The 1D convergence proof is self-contained and, if correct, appears to be one of the first rigorous analyses of a correction-based Cartesian grid method for a moving-domain parabolic problem with Robin data. The design choice to keep the bulk operator unchanged is practically valuable, as it permits geometric multigrid and matrix-free GMRES and avoids cut-cell geometry. The paper is transparent that the 2D convergence proof is open, and the numerical evidence is generally consistent with the claimed rates. The main limitations are a technical gap in the application of Theorem 4.7 to the auxiliary function in Lemma 4.12, the absence of nonzero-alpha 2D manufactured tests, and the reliance on a self-cited 6x6 collocation solvability result. These are addressable without changing the central method.","major_comments":[{"comment":"Equation (4.104) is not a special case of the canonical reduced equation (4.30) because (4.30) also contains the term (h/Q^{n-1}) R^n Z^{n-1} \\tilde g^{n-1} when \\tilde g = \\hat g. As written, Theorem 4.7 does not directly apply to (4.104). The gap is repairable: either extend Theorem 4.7 to the E-only source (the comparison proof is unchanged) or define \\hat b with the full source and absorb the extra O(h^3) crossing term into \\bar e r2. Please make this explicit. I do not agree with the scaling objection that the 1/h factor is dropped: Theorem 4.7 bounds the Robin source \\tilde g itself, and the sign in (4.105) gives the correct cancellation with Corollary 4.10.","section":"§4.5, Lemma 4.12 / Thm 4.7"},{"comment":"In both 2D manufactured tests the interface is transported by the prescribed velocity, so \\partial_t X = u and hence the dimensionless Robin coefficient alpha = (\\partial_t X - u)\\cdot n is identically zero. The 2D convergence studies therefore do not exercise the Robin transfer coefficient alpha, which is a central feature of the method. Please add a 2D manufactured test with nonzero alpha, or explain why the alpha=0 case is a sufficient validation of the Robin treatment in 2D.","section":"§5.2.1 and §5.2.2"},{"comment":"The 6x6 local collocation system is asserted to be square and solvable by reference to [37], a self-cited elliptic-interface paper. No condition number or accuracy analysis is given for the parabolic moving-interface setting, and nearest-marker patching is assumed accurate. Since the 2D method's accuracy and the claimed near-second-order rates depend on these local solves, please provide at least numerical conditioning evidence or an analysis for the present setting. The 1D theorem is unaffected because the correction coefficients are explicit.","section":"§3.3, Eqs. (3.25)-(3.26)"}],"minor_comments":[{"comment":"The title in the full text reads 'A CAR TESIAN GRID METHOD'; it should be 'A CARTESIAN GRID METHOD'.","section":"Title"},{"comment":"The notation C^n is used both for the vector of exact nodal values and for the correction function; the bold/non-bold distinction is easy to miss. Consider renaming one of the two objects.","section":"§4.2, Eq. (4.9)"},{"comment":"The convergence plots in Figures 3, 6, and 8 would be easier to assess if the observed orders were tabulated or annotated; the text reports rates only for the 1D table.","section":"§5.2"},{"comment":"The statement that the overall cost is linear in N_h when GMRES counts are mesh-independent should explicitly cite the figures and report representative iteration counts, rather than referring to 'below'.","section":"§3.4"},{"comment":"The extension to interfaces with finitely many direction changes is stated as a remark; since it is used in the numerical section's monotone test, it would be cleaner to state it as a corollary with a one-sentence proof sketch.","section":"Remark 4.14"},{"comment":"The phrase 'the system is square; see [37]' should state the exact proposition in [37] and whether it covers the time-derivative term present in the parabolic collocation system.","section":"§3.3, p. 8"}],"recommendation":"major_revision","confidential_remarks":"The technical gap in Lemma 4.12 is likely repairable, but it should be fixed before publication because it sits at the core of the only proved convergence theorem. The 2D validation gap (alpha=0) and the reliance on the self-cited [37] for the 6x6 systems are also worth addressing. The novelty claim of being the first convergence proof for this class is plausible, but the authors should double-check against the closest related analyses, especially [8] and [16]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this is a real extension of correction-function Cartesian methods to moving-domain Robin problems, and the 1D convergence proof is the actual payload. I checked the stress-test objection about Lemma 4.12 and I don't think it lands. The factor 1/h in (4.104) is part of the discrete boundary operator; Theorem 4.7 bounds the solution by sup |\\hat g|, not by the vector RHS. The sign of \\hat g is also consistent with cancellating the O(\\tau h) term in Corollary 4.10. So the O(\\tau+h^2) theorem appears to hold, though a referee should double-check the comparison argument.\n\nWhat's new: the reformulation of the moving-domain Robin problem as an interface problem with an unknown density, local polynomial corrections obtained from a 6x6 collocation system, and a fully discrete 1D analysis via a discrete maximum principle. That analysis is rare for correction-based methods, and it's done carefully. The numerics are honest: 1D convergence matches the theorem, and 2D manufactured tests show near second order.\n\nSoft spots, in proportion: the 2D method is the weak half. The solvability of the local 6x6 systems is delegated to a self-cited paper [37] with no analysis here, and the 2D tests never exercise a nonzero Robin coefficient \\alpha — in both manufactured cases the interface moves with the fluid, so \\alpha = 0 and the Robin condition degenerates to Neumann. That's a real gap for a paper advertising Robin conditions. Also missing: any baseline comparison with existing moving-domain methods (cut-cell, integral-equation). These are addressable, not fatal.\n\nWho it's for: specialists in Cartesian / immersed interface methods for parabolic problems, and anyone extending correction-function schemes. It deserves a serious referee; the 1D theorem is worth checking in detail, and the 2D concerns are exactly what a referee should probe.\n\nRecommendation: conditional acceptance — have the authors either prove or clearly state the 2D collocation assumption, and add at least one 2D test with \\alpha \\neq 0. If the journal treats the 1D analysis as the main result, the rest is presentable as supporting evidence.","headline":"Solid 1D convergence proof for a correction-based moving-domain Robin method; the 2D half leans on an unanalyzed collocation assumption and never tests nonzero Robin data.","tokens_in":28695,"tokens_out":14502,"would_cite":true,"duration_ms":124007,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M06","65M12","65M85","35K20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper develops a Cartesian grid method for moving-domain advection–diffusion equations with Robin boundary conditions, proves O(τ + h²) convergence in one dimension, and shows comparable accuracy in two dimensions.","keywords":["Cartesian grid method","moving domain","Robin boundary condition","correction function","interface density","advection-diffusion","finite difference","convergence analysis"],"falsifier":"Compute the determinant and condition number of the 6×6 collocation matrix from (3.25)–(3.26) at every marker for the star-shaped example over the whole run. If any determinant is zero, or the condition number grows like h^{-k} for k≥1 while errors still appear second-order, the 2D robustness claim is unsupported. Alternatively, run a 2D manufactured-solution test with the interface deliberately arranged so that collocation points X(s_k+jηΔs) become nearly degenerate (e.g., collinear); if the scheme still gives second-order, the local-system analysis is incomplete.","tokens_in":27719,"feed_emoji":"🧮","tokens_out":7407,"duration_ms":65130,"temperature":0.7,"pith_summary":"The paper claims that moving-domain advection–diffusion problems with Robin boundary conditions can be solved on a fixed Cartesian grid, with no remeshing and no cut cells, by introducing an unknown interface density and local polynomial corrections that enter only the right-hand side of the discrete equations. For the one-dimensional version of the scheme, the paper proves a fully discrete error bound of O(τ + h²) — first order in time, second order in space — under the mesh constraints 1/2 h² ≤ τ < h/max|γ̇| and monotone (or finitely piecewise-monotone) interface motion. In two dimensions, numerical experiments with manufactured solutions and an active transport problem show near-second-order convergence and GMRES iteration counts that are essentially independent of mesh size, implying a solve cost linear in the number of bulk unknowns. The central mechanism is a correction function C computed from a local 6×6 collocation problem at each interface marker, which supplies jump information so the bulk operator stays the standard Cartesian finite-difference matrix.","feed_headline":"Fixed-grid method proves second-order for moving-boundary diffusion","feed_subtitle":"A correction-based Cartesian method avoids remeshing and cut cells, with linear-cost solves and a proven error bound in 1D.","key_machinery":"The key object is the correction function C = c⁺ − c⁻ in a narrow band around the interface. In 2D it is represented as a local quadratic polynomial (six coefficients) at each marker, determined by a 6×6 collocation system imposing the PDE, the Dirichlet condition C = 0 on the interface, and ∂nC = ψ; the correction Q entering the discrete equation is the commutator [L_{τ,h}, χ]C, so it is nonzero only at irregular stencils. In 1D the same idea reduces to explicit coefficients (4.14)–(4.15). The boundary condition is imposed by Lagrange interpolation of {c} and {∇c} at interface points, closing the system (3.27); the analysis then rests on a discrete maximum principle and an M-matrix argument","core_discovery":"The paper's core discovery is that a moving-domain Robin boundary condition can be enforced through an unknown interface density ψ, converting the problem into an interface problem on a fixed box with a jump in the normal derivative. The discrete scheme then needs only local corrections near the interface: a quadratic correction function C that satisfies, in a local collocation sense, the PDE, the zero-value condition, and ∂nC = ψ. Because these corrections enter only the right-hand side, the bulk coefficient matrix remains the standard cell-centered Cartesian finite-difference operator, solvable by geometric multigrid; eliminating correction coefficients and bulk unknowns leaves a boundary-","pith_inferences":["The unsupported part is the 2D local collocation: the paper asserts the 6×6 systems are square by citing prior work, but gives no analysis of their solvability, conditioning, or approximation error. A robustness fix would be to use least squares with more collocation points (Nc > 6) or to verify full-rank of the Vandermonde matrix at run time.","The nearest-marker patching rule is the simplest possible; a smoother patching (e.g., weighted average of overlapping local polynomials) might improve accuracy on strongly curved interfaces, though it would complicate the right-hand-side-only structure.","The proof's monotone-motion assumption is likely removable by a more careful summation bound for interfaces that cross the same grid node multiple times, but the error constant would then depend on the total variation of γ.","A 3D extension would require a much larger local polynomial space and a more elaborate collocation set, so the 2D-to-3D step is non-trivial."],"forward_implications":["If the method is correct, moving-boundary problems with Robin conditions can be advanced on fixed grids at a cost linear in the number of bulk unknowns, with only right-hand-side corrections and a small boundary-only GMRES solve.","The 1D proof provides a template for showing that large, localized consistency errors at a moving interface need not degrade the global convergence rate; analogous arguments may apply to other correction-based immersed-interface schemes.","The interface-density formulation connects the method to boundary integral theory, so tools from discrete potential theory could be used to analyze the well-conditioning of the reduced system (3.30).","Practitioners get an explicit stability window, τ between 1/2 h² and h/max interface speed, for the backward-Euler + correction scheme.","For the motivating osmotic cell-migration application, the method can simulate coupled active transport with mass conservation errors that shrink under refinement, as demonstrated by the conservation diagnostics."],"fun_headline_variants":["Moving-domain Robin conditions enforced by an interface density","No cut cells, no remeshing: a linear-cost Cartesian solver","Proven second-order convergence for moving-boundary diffusion","Matrix-free GMRES with mesh-independent iterations for moving domains","Interface problem on a box solves moving-domain advection-diffusion"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The two-dimensional method's accuracy depends on the unproven solvability and accuracy of the 6×6 collocation systems that build the correction function at each interface marker; if those systems become singular or ill-conditioned for some interface geometry, the claimed second-order convergence in 2D collapses.","fun_headline_variants_meta":{"raw":{"variants":["Moving-domain Robin conditions enforced by an interface density","No cut cells, no remeshing: a linear-cost Cartesian solver","Proven second-order convergence for moving-boundary diffusion","Matrix-free GMRES with mesh-independent iterations for moving domains","Interface problem on a box solves moving-domain advection-diffusion"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000949,"raw_usage":{"total_tokens":3859,"prompt_tokens":691,"completion_tokens":3168,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":435,"completion_tokens_details":{"reasoning_tokens":3098}},"tokens_in":435,"tokens_out":3168,"duration_ms":19240,"temperature":1.0,"reasoning_tokens":3098,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T00:05:47.499699+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the determinant and condition number of the 6×6 collocation matrix from (3.25)–(3.26) at every marker for the star-shaped example over the whole run. If any determinant is zero, or the condition number grows like h^{-k} for k≥1 while errors still appear second-order, the 2D robustness claim is unsupported. Alternatively, run a 2D manufactured-solution test with the interface deliberately arranged so that collocation points X(s_k+jηΔs) become nearly degenerate (e.g., collinear); if the scheme still gives second-order, the local-system analysis is incomplete.","supporting_citations":[],"review_version":1}