{"id":"95884be8-cb3c-4237-b7ed-120db53d3ed7","arxiv_id":"2607.21295","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A MAC-staggered lattice Green's function pair yields an unfitted boundary algebraic equation method for 2D Stokes flow with second-order convergence and Calderón preconditioning.","lead":"The paper develops a numerical method for slow viscous (Stokes) flow around arbitrary obstacles, using a regular grid plus specially constructed lattice Green's functions so only the boundary enters the equations. Tests show second-order accuracy, discrete incompressibility at solver precision, and fast convergence of the iterative solver.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Rank completion in Eq. (4.17) is asserted, not proven; if nΓ fails to represent the true algebraic hydrostatic mode, Ac remains singular and the entire solve fails, so the central claim needs a direct nullspace/nonsingularity check.","rationale":"The reader identified the sampled-normal rank completion as the weakest assumption, and I agree. The central claim—second-order accurate, robust Stokes solutions on unfitted MAC grids—depends on solving a square boundary system. If that system is singular, the method fails outright, so the rank completion is more load-bearing than the Calderón preconditioner (which affects only efficiency) or the slightly uneven pressure convergence rates (an overclaim but not a correctness failure). The Fourier derivation of the LGF pair is well supported and the numerical tests are extensive, but they cannot substitute for a verification that the algebraic nullspace of A is exactly the K-dimensional hydrostatic space and that nΓ^k is a faithful basis for it. The proposed SVD-based check directly tests this: it computes the actual nullspace and measures whether the sampled-normal update lifts it. If the check passes on a diverse set of geometries, the concern is resolved; if it fails, the method would need a different completion strategy. Since the concern is real but addressable and the reader's CONDITIONAL verdict already reflects it, no change to the verdict is needed.","tokens_in":24219,"tokens_out":12908,"duration_ms":144088,"concrete_test":"For an existing test geometry (e.g., the three-obstacle exterior case) and one new irregular multi-object geometry not in the paper, assemble the matrix A (from the authors' code or by reimplementation), compute its SVD, and check: (i) nullity(A) equals the number of objects K; (ii) for each null vector z, the principal angle between span(NΓ) and ker(A) is O(h) or at least not near 90°, equivalently the smallest singular value of NΓ^T Z is well above machine precision; (iii) the smallest singular value of Ac = A + τ NΓ NΓ^T is O(1) relative to the non-null spectrum and not at the rounding level. If (iii) fails for any geometry, the sampled-normal rank completion does not robustly remove the hydrostatic nullspace.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The method's correctness hinges on the rank-one completion in Eq. (4.17): the boundary system Aq = g is singular with an algebraic hydrostatic nullspace Zh = ∇hχ|γ− (Theorem 3.2), and the paper removes it by adding τ nΓ nΓ^T, where nΓ is the outward normal sampled at boundary intersection points. Remark 4.1 asserts that nΓ is 'the cut-point representation' of the exact discrete null vector zh 'up to discretization and scaling', but gives no proof and no general numerical check. This is not a minor technicality: if nΓ is not sufficiently aligned with every null direction of A (and, for multiple objects, if the matrix NΓ^T Z is singular), then the completed operator Ac remains singular or nearly singular, and the linear system has no unique solution. The relationship is delicate because zh is a discrete gradient of a cell-centered indicator, supported on staggered grid nodes with magnitude 1/h along grid directions, whereas nΓ is a continuous normal vector sampled at cut points; the two can differ substantially for boundaries nearly tangent to grid lines or for non-convex shapes with multiple intersections on the same grid line. The numerical experiments in Section 9 demonstrate that the completion worked for the tested geometries, but they do not establish the general claim. The paper provides no analysis of the nullspace of A itself (only of S−), so additional non-hydrostatic null vectors introduced by the interpolation matrices Φ± are also not ruled out. This is the single most load-bearing concern because it affects existence and uniqueness of the discrete solution, not just conditioning or iteration counts.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an unfitted boundary algebraic equation (BAE) method for 2D steady Stokes flow on a MAC grid. It explicitly constructs free-space velocity and pressure lattice Green's functions for the discrete MAC Stokes operator from Laplace/biharmonic lattice Green's functions (Theorem 2.1), represents homogeneous fields by sources on thin staggered boundary layers, enforces Dirichlet data by local cut-point interpolation, removes hydrostatic null modes by sampled-normal rank updates, and accelerates/solves the resulting dense boundary system with FFT convolutions and a componentwise discrete Calderón preconditioner built from the scalar Laplace LGF. Numerical experiments cover interior, multiply connected, narrow-gap, body-forced, corner-eddy, and exterior configurations, reporting roughly second-order velocity convergence, small discrete divergence, and reduced condition numbers.","tokens_in":24692,"tokens_out":8151,"duration_ms":91042,"significance":"If the construction is fully rigorous, the paper offers a valuable new discretize-then-represent boundary algebraic framework for Stokes flow that avoids singular quadrature and artificial outer boundary conditions, with an explicit, reusable lattice Green's function pair and a promising preconditioning strategy. The derivation of the MAC Stokes LGF is clean and self-contained, and the numerical evidence is extensive and consistently supportive: second-order velocity convergence across many geometries, divergence at solver accuracy, and substantial condition-number reduction. The main value lies in the combination of staggered-grid discrete potential theory with unfitted boundary closure. However, two analytical pillars of the method are asserted rather than proved: the sampled-normal rank completion that removes the hydrostatic nullspace, and the discrete Calderón identity underlying the preconditioner. These gaps currently limit the strength of the general algorithmic claims, even though the numerical results are encouraging.","major_comments":[{"comment":"The rank completion is the keystone of the boundary closure, but it is asserted rather than proved. Theorem 3.2 only identifies ker S−; it says nothing about ker A = Φ+S+ + Φ−S−, and Remark 4.1 gives no argument that the sampled normal nΓ is aligned with the algebraic hydrostatic mode zh = ∇hχ|γ− for arbitrary cut configurations. For K objects one additionally needs NΓᵀZ to be nonsingular. If nΓ misses a null direction (e.g., a boundary almost tangent to a grid line, or a non-convex shape with multiple intersections on one grid line), Ac in (4.17)/(4.19) remains singular and the solve is ill-posed. The experiments in Section 9 show the completion worked for the tested geometries but do not establish the general claim. Please add (i) a direct numerical check of the smallest singular value/algebraic nullspace of A and Ac for a range of unfitted geometries, including near-tangent cuts and m","section":"Section 4, Eq. (4.17), Remark 4.1"},{"comment":"The energy argument assumes that the regularized LGF potential generated by a force-balanced density has finite discrete energy and that the only finite-energy homogeneous discrete Stokes field with zero trace on γ− is zero. For 2D free-space Stokes, a force dipole has velocity decaying as 1/r, so the continuous energy integral is logarithmically divergent; whether the chosen kernel regularization/gauge restores finite energy on the lattice is not shown. Moreover the boundary system is not actually constrained to the force-balanced class during the solve, so the relevant object is the nullspace of A, which includes the interpolation operators Φ±; non-hydrostatic null vectors introduced by the closure are not ruled out. Please prove or carefully state these spectral assumptions, and provide a numerical nullspace diagnostic for A.","section":"Section 3.4, Lemma 3.1 and Theorem 3.2"},{"comment":"The discrete Calderón identity Vα(Jα+Wα) = −Kα(Iα+Kα) is stated as a fact for the unfitted staggered boundary layer with incidence scaling Jα. The cited source [29] establishes related identities for fitted lattice boundaries; the unfitted case with variable incidence is new and no proof is supplied. Since the identity underlies the claimed mesh-independent preconditioning, the paper should either prove it under explicit assumptions or present it as a heuristic supported by the numerical conditioning data. At minimum, state the precise sense in which the identity holds (exactly, up to O(h), or only in a spectral average sense).","section":"Section 5, Eq. (5.6)"}],"minor_comments":[{"comment":"The abstract states 'second-order velocity and pressure convergence', but Tables 3 and 11 show pressure rates around 1.55–1.80 in the Taylor–Couette and single-obstacle exterior cases. I suggest qualifying the claim as 'approximately second-order velocity and close to second-order pressure in most configurations'.","section":"Abstract and Section 9.3"},{"comment":"The Moffatt-eddy ratios are compared to the asymptotic value 2.10 on a single grid. A brief refinement study of the eddy-center locations would substantiate the claim that the method resolves the asymptotic scaling rather than producing the right value by coincidence.","section":"Section 9.1.5"},{"comment":"The text says the total runtime scaling is 'consistent with near-optimal O(N² log N)', but the reported total times approximately quadruple per refinement, which is also consistent with O(N²). The paper should state which component dominates the log factor or present a more detailed scaling breakdown.","section":"Section 9.1.2, Figure 7"},{"comment":"For body forcing that is not compactly supported (e.g., the trigonometric forcing in Section 9.1.6), the FFT-based convolution truncates the infinite source distribution after padding. The paper should state the implicit decay/truncation assumption and note that sufficient padding is needed to control aliasing for non-decaying data.","section":"Section 7"}],"recommendation":"major_revision","confidential_remarks":"The numerical work appears careful and reproducible, and the LGF derivation is a solid contribution. The main barrier to acceptance is the gap between the algebraic claims in Sections 4 and 5 and their justification; the rank-completion issue in particular is load-bearing and should be addressed with either a proof or a systematic numerical nullspace/conditioning check. The self-citation is moderate and appropriate given the direct lineage from the authors' scalar BAE work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this paper really does construct something new—an explicit free-space velocity–pressure lattice Green's function pair for the MAC Stokes operator (Theorem 2.1) and uses it as the kernel of an unfitted boundary algebraic equation. That is a real advance over prior scalar-Laplace LGF projection solvers and over the authors' own scalar BAE work. The numerical side is thorough: second-order velocity convergence across interior, exterior, multiply connected, narrow-gap, body-forced, and Moffatt tests; divergence at solver accuracy; and a Calderón preconditioner that reduces condition numbers by orders of magnitude and is nearly mesh-independent for exterior problems. The Moffatt eddy scale ratios are compared to the theoretical 2.10 and match to within about 0.7%. Credit where due: the Fourier derivation is clean, and the experiments are broad enough that the convergence claims are credible.\n\nThe soft spot is the one the stress-test flags, and it is load-bearing. Section 4 completes the singular boundary system by adding τ nΓ nΓᵀ, with nΓ the outward normal sampled at cut points. Remark 4.1 asserts this represents the discrete hydrostatic null vector z_h 'up to discretization and scaling.' That is not a proof. If nΓ is not sufficiently aligned with each hydrostatic null mode—or, for multiple objects, if NΓᵀ Z is singular—the completed A_c remains singular and the whole solve is not well-posed. The paper never analyzes the nullspace of the full A, only of S⁻, so there is also no accounting for extra null vectors that the interpolation matrices could introduce. This is a correctness issue, not an efficiency one. The experiments suggest the completion worked for the tested geometries, but the general claim needs either a nonsingularity/nullspace argument or a systematic numerical check.\n\nTwo smaller issues: the Calderón identity (5.6) is imported from a scalar lattice context, and the J_α incidence scaling is described in a single sentence; the numerics support it, but the analysis is thin. Also, no code or data is released, which makes independent verification slower. Minor point: pressure convergence is more variable in two tests, though still monotone.\n\nWho is this for: anyone working on boundary integral or unfitted methods for Stokes and incompressible flow. It merits a serious refereeing; the referee should be asked to scrutinize Remark 4.1 and the preconditioner derivation. I would accept for review and likely endorse conditional acceptance after that gap is closed.","headline":"A genuinely new discretize-then-represent Stokes BAE with explicit MAC lattice Green's functions and strong numerics, but the rank-completion step that guarantees solvability is asserted rather than proved.","tokens_in":25079,"tokens_out":2773,"would_cite":true,"duration_ms":32128,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N38","65N06","76D07"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper constructs explicit lattice Green's functions for the 2D staggered MAC Stokes operator and shows a boundary-only algebraic system solves irregular-geometry Stokes flow with second-order accuracy.","keywords":["Stokes equations","lattice Green's function","boundary algebraic equation","discrete potential theory","MAC scheme","Calderón preconditioning","unfitted mesh","Moffatt eddies"],"falsifier":"Compute the exact nullspace of the uncompleted boundary matrix A for a non-convex or multiple-obstacle geometry at several grid spacings, and compare its null vectors to the sampled-normal vector nΓ (and per-object normals). If any null vector has substantial component away from nΓ, or if the completed system shows a near-zero singular value that does not shrink under refinement, the rank-completion premise fails.","tokens_in":24132,"feed_emoji":"🌊","tokens_out":3845,"duration_ms":35779,"temperature":0.7,"pith_summary":"The paper establishes that steady 2D Stokes flow on a staggered MAC grid can be solved as a fully discrete boundary algebraic equation, with no fitted mesh, no singular quadrature, and no artificial far-field boundary conditions. It builds explicit velocity and pressure lattice Green's functions from Laplace and biharmonic lattice Green's functions, then represents homogeneous incompressible fields by sources on thin staggered boundary layers, with Dirichlet data imposed at cut points. Sampled-normal rank updates remove the hydrostatic pressure-jump nullspaces, one per disconnected obstacle, and a componentwise Calderón preconditioner built from the scalar Laplace kernel makes the dense boundary system well conditioned. Numerical tests show second-order velocity and pressure convergence, discrete divergence at solver accuracy, and near mesh-independent conditioning for exterior flows.","feed_headline":"Lattice kernels give second-order Stokes flow on irregular domains","feed_subtitle":"No fitted mesh, no singular quadrature: a boundary-only system keeps discrete divergence at solver accuracy.","key_machinery":"The key machinery is the MAC Stokes lattice Green's function pair (S,P), derived in Theorem 2.1 from the Laplace lattice Green's function G and a gauge-fixed biharmonic representative H via the Duffin–Shelly identity. S is a 2x2 tensor of difference operators applied to H; P is a gradient of G. These translation-invariant kernels invert the MAC Stokes operator and supply both homogeneous layer potentials and volume potentials. Around each kernel sit three supporting devices: cut-point interpolation with piecewise-linear hat functions and optional extrapolation to extra exterior nodes; sampled-normal rank updates that project out the hydrostatic nullspace of the single-layer matrix; and a com","core_discovery":"The central discovery is a translation-invariant free-space pair of lattice Green's functions for the coupled steady Stokes operator on the staggered MAC grid: the velocity kernel S and pressure kernel P, constructed in closed form from regularized Laplace and biharmonic lattice Green's functions. These kernels satisfy the discrete momentum equations and the discrete divergence constraint, so convolution with layer sources on thin exterior boundary layers produces homogeneous incompressible fields. Dirichlet data are imposed by local tensor-product hat interpolation at cut points, and the resulting boundary-density equation is completed by a sampled-normal rank update that removes one hydros","pith_inferences":["If the sampled-normal rank completion faithfully represents the exact discrete hydrostatic null vector (a claim the paper asserts without a rigorous proof), the method should extend to arbitrarily many disjoint obstacles by adding one rank-one update per object; a careful numerical test comparing the sampled normal to the exact algebraic null vector for a non-convex obstacle would settle this.","The explicit Stokes lattice Green's function pair opens a natural route to a discrete double-layer formulation, which the paper lists as future work; such a formulation could reduce the boundary system's size and improve conditioning in narrow gaps.","The same discretize-then-represent route should transfer to 3D MAC Stokes and to linear elasticity, since the derivation only needs the Laplace and biharmonic lattice Green's functions and the staggered-grid projection; the main challenges would be larger boundary-layer cardinality and a multidirectional sampled-normal completion.","For time-dependent Navier–Stokes, the volume-potential/homogeneous-correction split suggests a semi-implicit scheme where nonlinear terms are treated as known volume sources at each step, reusing the same boundary operator."],"forward_implications":["Second-order velocity and pressure convergence is demonstrated across smooth, multiply connected, narrow-gap, body-forced, and exterior configurations, so the method is accurate without boundary-fitted meshes.","The maximum discrete divergence is bounded between 10^-12 and 10^-7 in all tests, confirming that the unfitted closure preserves the MAC incompressibility constraint to solver accuracy.","The Calderón preconditioner reduces the condition number by orders of magnitude and gives nearly mesh-independent conditioning for exterior flows, with GMRES iteration counts in the tens to low hundreds.","The recovered Moffatt-eddy scale ratios (2.105, 2.113, 2.095 versus the asymptotic 2.10) show that the method resolves multiscale corner structures.","Because the lattice kernels depend only on the background grid, they can be reused when geometry changes, and the boundary operator can be reused when only the body force changes, enabling efficient moving-boundary and time-dependent extensions."],"fun_headline_variants":["Boundary-only Stokes solver hits second order on cut grids","Lattice Green's functions: no mesh, no singular quadrature for Stokes","Calderón preconditioner speeds Stokes solve on irregular grids","Stokes flow on cut cells: boundary layers and Calderón to second order","Nearly mesh-independent conditioning for exterior Stokes flow via Calderón"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the sampled normal vector at cut points truly represents the discrete hydrostatic pressure-jump mode; if this proxy is wrong for some geometry, the rank-completed boundary system remains singular and the entire solve fails.","fun_headline_variants_meta":{"raw":{"variants":["Boundary-only Stokes solver hits second order on cut grids","Lattice Green's functions: no mesh, no singular quadrature for Stokes","Calderón preconditioner speeds Stokes solve on irregular grids","Stokes flow on cut cells: boundary layers and Calderón to second order","Nearly mesh-independent conditioning for exterior Stokes flow via Calderón"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1283,"prompt_tokens":757,"completion_tokens":526,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":501,"completion_tokens_details":{"reasoning_tokens":437}},"tokens_in":501,"tokens_out":526,"duration_ms":6222,"temperature":1.0,"reasoning_tokens":437,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T07:51:36.141024+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the exact nullspace of the uncompleted boundary matrix A for a non-convex or multiple-obstacle geometry at several grid spacings, and compare its null vectors to the sampled-normal vector nΓ (and per-object normals). If any null vector has substantial component away from nΓ, or if the completed system shows a near-zero singular value that does not shrink under refinement, the rank-completion premise fails.","supporting_citations":[],"review_version":1}