{"id":"9de7674a-4cc9-4995-bff2-939e8c74fb30","arxiv_id":"2512.10473","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A POD-based reduced-order model for the unsteady Stokes equations using a BDF2 incremental projection scheme is claimed to be second-order accurate, but the printed numerical validation and reduced system contain errors.","lead":"This paper builds a reduced-order model for the unsteady Stokes equations by combining a second-order incremental projection scheme with proper orthogonal decomposition, and derives stability and error estimates for it. The numerical validation, however, contains several errors—including a non-divergence-free manufactured solution—so the experimental support is currently invalid.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.2's central error bound rests on an unverified saturation property α<3/8 for POD spaces; no proof is given and the numerics use L2-POD, so the claim is unsupported.","rationale":"The reader's weakest-assumption analysis identifies exactly the saturation property as the load-bearing condition for Theorem 3.2, and I agree. The theorem's error estimate is conditional on α<3/8 for the pair (U_r, ∇Q_r), but the paper supplies no proof, no numerical estimate, and no argument that POD spaces inherit such an angle condition. In fact, the numerical experiments deliberately use L2-based POD for both fields, which is a different construction from the H1-based POD analyzed in Section 3.1, so the experiments cannot be read as verifying the hypothesis. This is not a minor technical gap: the bound in Eq. (3.26) fails to follow if α≥3/8, and the entire error analysis collapses. Other issues noted by the reader—the non-divergence-free manufactured solution in (4.1), the sign/transpose inconsistency in the algebraic system (3.5), and the unspecified initial pressure data—are real and independently support rejection, but the saturation assumption is the most directly load-bearing for the central theoretical claim. Because the paper as written does not establish this assumption and the numerical section does not test it, the verdict remains REJECT with no adjustment needed.","tokens_in":21195,"tokens_out":15546,"duration_ms":139571,"concrete_test":"Take the Section 4.1 snapshots and regenerate the reduced spaces strictly following Section 3.1: build velocity POD modes φ_i from the H1 correlation matrix (∇u_i,∇u_j) and pressure modes ψ_i from (∇p_i,∇p_j). Then compute the saturation constant α = sup_{y∈U_r, z∈span{∇ψ_1,...,∇ψ_r}} |(y,z)_L2| / (||y||_L2 ||z||_L2), with the supremum taken over nonzero elements and verifying that U_r ∩ span{∇ψ_j} = {0}. If the computed α is ≥3/8, or if the intersection is nontrivial, then the hypothesis of Theorem 3.2 is violated for this test case and the stated error bound cannot be invoked. If instead α<3/8, the theorem may apply, but the printed numerical experiments would still need to be repeated with H1-based POD before they count as confirmation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 3.2: the ROM error is bounded by a constant times the POD projection error, provided the saturation constant α of Lemma 3.1 satisfies α<3/8. The saturation property (Definition 3.1, Eq. (3.23)) is assumed, not proved, for the specific POD reduced spaces U_r and ∇Q_r. In fact, Definition 3.1 requires the intersection of Y and Z to be {0}, but the paper never checks this for the POD-generated spaces, nor does it estimate α. The proof uses the saturation hypothesis in a load-bearing way: e.g., the A6 estimate (Eq. (3.41)) bounds a cross term between U_r and ∇Q_r via α, and the later absorbing steps leading to Eq. (3.46) and the Gronwall argument require the coefficient 1−17τ/4−α to remain positive; with the stated condition τ<4(1−8α/3)/17, α<3/8 is essential. Without a guarantee that α<3/8 for the H1-POD spaces, the theorem does not apply. The numerical section does not rescue this: Section 4.1.1 and Section 4.2.1 explicitly state that POD modes are generated in the L2 norm for both velocity and pressure, while the theoretical POD in Section 3.1 uses the H1 seminorm (∇ϕ_i,∇ϕ_j) and (∇ψ_i,∇ψ_j). Thus the experiments test a different reduced-space construction and cannot provide empirical evidence for the saturation hypothesis used in Theorem 3.2.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a POD-based reduced-order model for the unsteady Stokes equations built on a BDF2 incremental projection scheme. It presents semi-discrete and fully discrete formulations, a stability estimate (Theorem 3.1), and an a priori error estimate (Theorem 3.2) in which the ROM error is bounded by POD truncation eigenvalues up to a factor depending on a saturation constant alpha < 3/8. Numerical experiments with a manufactured solution and a lid-driven cavity are reported. The main claim is that the reduced velocity and pressure are accurate with few modes and that the scheme preserves second-order accuracy in time.","tokens_in":21587,"tokens_out":14755,"duration_ms":140736,"significance":"If the main theorem were fully established, the paper would be a useful contribution: it combines a popular pressure-correction time-splitting scheme with POD, and the claimed bound by POD truncation error is an attractive and practically relevant statement. The paper also attempts a complete stability analysis. However, the numerical validation is invalid because the manufactured solution is not divergence-free, the printed reduced algebraic system is inconsistent with the Galerkin weak form, the experiments use L2-POD while the theory uses H1-POD, and the central error estimate rests on an unverified saturation hypothesis and contains gaps in the Gronwall argument. As it stands, the central claims are not supported.","major_comments":[{"comment":"The manufactured solution is not divergence-free. Both components are printed as u=v=cos(t) pi sin^2(pi x) sin(2 pi y), whose divergence is 2 pi^2 cos(t) sin(pi x)(cos(pi x) sin(2 pi y)+sin(pi x) cos(2 pi y)), not identically zero. It therefore violates (2.1b) and is not a solution of the Stokes problem. Consequently the convergence rates in Table 1 and the ROM errors in Figures 2-3 do not measure errors against a valid exact solution. A divergence-free manufactured solution is required to validate the scheme or the theory.","section":"Section 4.1, Eq. (4.1)"},{"comment":"The reduced algebraic system is inconsistent with the stated weak form. The pressure term in (3.4a) is -1/3 (p, div v_r). With P_ij=(psi_i, div phi_j), this contributes -1/3 P^T b, not +1/3 P b as printed in (3.5a). Similarly, (div u^{n+1}, psi_i) with the stated definition of P gives P a^{n+1}, so (3.5b) should not contain P^T. As printed, the system does not represent the Galerkin equations (3.4). Any implementation following (3.5) solves a different ROM, so the numerical results do not validate the printed scheme.","section":"Section 3.2, Eqs. (3.4a)-(3.5b)"},{"comment":"The theory and experiments use different POD inner products. Section 3.1 constructs POD bases with X=H^1-seminorm and Y=H^1-seminorm, with orthonormality (nabla phi_i, nabla phi_j)=delta_ij; but Sections 4.1.1 and 4.2.1 explicitly state that 'POD modes are generated in L2-norm for velocity and pressure'. The POD projection error estimates and the saturation property in Theorem 3.2 are formulated for the H1-POD spaces, so the experiments do not test the spaces appearing in the theory. The paper must either adapt the analysis to L2-POD or run the numerical tests with H1-POD.","section":"Sections 3.1 and 4.1.1/4.2.1"},{"comment":"The saturation property is assumed, not verified, for the specific reduced spaces U_r and span{nabla psi_1,...,nabla psi_r}. Definition 3.1 requires Y cap Z = {0} and (3.23), and Theorem 3.2 assumes the saturation constant satisfies alpha < 3/8. The paper proves no such property for POD bases and gives no computational estimate of alpha. This is load-bearing: the absorption step leading to (3.46) requires positivity of 1-17 tau/4-alpha, and the Gronwall coefficient depends on alpha. Without a proof, a verifiable criterion, or an estimate of alpha for the actual POD spaces, the advertised error bound is conditional on an unestablished geometric hypothesis.","section":"Definition 3.1, Lemma 3.1, Theorem 3.2"},{"comment":"Two central steps in the proof are not justified. First, the estimate (3.41) for A6 is not a consequence of Lemma 3.1: A6 is an L2 pairing between an element of U_r and an element of span{nabla Q_r}, whereas Lemma 3.1 controls the X-inner product. If X=H^1, the controlled inner product is (nabla . , nabla .); if X=L^2, the POD construction is different. No inverse or norm-equivalence inequality is stated. Second, the discrete Gronwall step to (3.52) is inconsistent with (3.51): the coefficient on the energy-type sum is (7 tau + 4 alpha)/4, so discrete Gronwall gives a factor (1+(7 tau+4 alpha)/4)^{N_t} ~ exp((alpha+7 tau/4)T/tau), not exp((7 tau+4 alpha)T/2). The exponent in (3.52) has the wrong scaling in T and tau, which affects the validity of the stated error bound.","section":"Theorem 3.2 proof, Eqs. (3.41), (3.51), (3.52)"}],"minor_comments":[{"comment":"The lid-driven cavity test solves the nonlinear Navier-Stokes equations with the extrapolated convection term ((2 u^n - u^{n-1}) . grad) u^{n+1}, while the stability and error analyses in Section 3 are for the linear Stokes problem. These results should be labeled as heuristic and outside the scope of Theorem 3.2.","section":"Section 4.2"},{"comment":"The summation indices in the POD basis formulas run from k=0 to N_t+1, but the snapshots are defined for indices 1 to N_t+1; u^0 is not defined. The index range should be corrected.","section":"Section 3.1"},{"comment":"The statement that 81 snapshots are collected from storing every fourth FOM simulation in [0.21,1] appears inconsistent with tau=1e-2; the number of stored snapshots and the time interval should be reconciled.","section":"Section 4.1.1"},{"comment":"The abstract claims numerical validation of second-order convergence in time, but Table 1 tests the full-order scheme, not the ROM; the ROM experiments show errors versus mode number rather than time-step convergence. The text should distinguish these.","section":"Abstract and Conclusion"}],"recommendation":"reject","confidential_remarks":"The paper has several load-bearing issues that cannot be fixed by minor revision: the manufactured solution is not divergence-free, the printed reduced system is inconsistent with the weak form, the numerical POD uses a different norm than the theory, and the main theorem rests on an unverified saturation hypothesis with an invalid Gronwall step. The authors would need to substantially rework both the analysis and the numerical validation before the claims could be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a legitimate extension of the authors' earlier Goda-based ROM to BDF2 incremental projection, with a stability proof and an error estimate worth reading. But as it stands, the numerical section is invalid, the reduced system as printed has a sign and dimension error, and the central error bound rests on an unverified saturation assumption. Don't take the numerics as evidence.\n\nWhat's new: the ROM construction and error analysis for this specific time-splitting scheme. The stability argument (Theorem 3.1) follows the usual three-term recursion, and Theorem 3.2 gives the expected bound: ROM error controlled by POD projection error, up to an exponential factor. That is not a breakthrough, but it is a reasonable extension within a useful line of work.\n\nWhere it falls short. First, the manufactured solution (4.1) is not divergence-free. Both components are given by the same expression, and ∂x u + ∂y v is not zero. Since the scheme enforces incompressibility, the 'exact' solution cannot be approximated in the expected sense; the convergence table is therefore meaningless. Second, the printed reduced system (3.5a) has a wrong sign on the pressure term and a dimension mismatch: the matrix P is Np×Nu, so the term should be -1/3 P^T(7bn - 5bn-1 + bn-2), not +1/3 P(...). Someone trying to implement from (3.5) would get stuck. Third, the theory assumes H1-POD (inner products with gradients), while the experiments explicitly generate modes in the L2 norm. So the numerical tests do not exercise the spaces the analysis requires. Fourth, Theorem 3.2 is conditional on a saturation constant α < 3/8 for the pair (U_r, ∇Q_r); the paper neither proves this nor offers a way to check it. That is a serious gap in the main theorem. Also minor: initial pressure data for the ROM are not specified, and the lid-driven cavity test adds convection while the analysis covers only Stokes.\n\nWho should read this: someone working on projection-based POD-ROMs who wants to see the BDF2-incremental variant analyzed; the proof technique may be reusable. But I would not cite the numerical results, and I would not recommend accepting the paper in this form. A serious referee should see it though—the core idea is meaningful and the defects are fixable.","headline":"A plausible but sloppy ROM paper: the BDF2-incremental-projection extension is real, but the numerical validation and the printed reduced system do not hold up.","tokens_in":22062,"tokens_out":4114,"would_cite":false,"duration_ms":40534,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N30","65M15","76D07"],"pacs":[],"model":"deepseek-v4-flash","headline":"A POD-based reduced-order incremental projection scheme for the Stokes equations is stable, and its velocity and pressure errors are bounded—up to an explicit exponential factor—by the POD projection error whenever the reduced velocity and","keywords":["reduced order model","proper orthogonal decomposition","incremental projection","Stokes equations","BDF2","error estimate","saturation property","incompressible flow"],"falsifier":"Compute α = sup_{y∈Ur, z∈∇Qr} |(y,z)_X| / (||y||_X ||z||_X) for the H1-based POD spaces built from the paper's snapshot ensembles; if for typical snapshot sets one finds α ≥ 3/8, the theorem's hypothesis is violated and the stated error bound is not guaranteed. Additionally, with α in hand, one can compare the measured ROM errors against the right-hand side of (3.26) at fixed ν and τ to test the bound directly.","tokens_in":21130,"feed_emoji":"🌊","tokens_out":6435,"duration_ms":69681,"temperature":0.7,"pith_summary":"The paper proposes a reduced-order model for incompressible flow that combines incremental projection time-splitting with proper orthogonal decomposition. It proves that the reduced scheme is stable and that, under a saturation condition on the reduced spaces, the velocity and pressure errors are controlled by the POD truncation error—the sum of the omitted POD eigenvalues. This means the reduced model's accuracy is essentially as good as the POD basis itself, and pressure is computed explicitly without needing an inf-sup condition. Numerical experiments confirm second-order time convergence and show that a few modes give very small errors.","feed_headline":"POD truncation error controls the reduced Stokes solve","feed_subtitle":"A stable projection ROM stays accurate whenever the reduced velocity and pressure-gradient spaces keep a safe angle.","key_machinery":"The central object is the saturation property (Definition 3.1): a pair of finite-dimensional subspaces (Y,Z) of a Hilbert space is saturated when ||y||+||z|| ≤ C||y+z|| for all y,z, equivalently |(y,z)| ≤ α||y||||z|| with α<1—an angle condition. The paper uses this with Y=Ur and Z=span{∇ψ_i} to control pressure–velocity cross terms. Around this, the method builds on an incremental (Goda-type) BDF2 projection scheme for the Stokes equations, H1-inner-product POD bases for both velocity and pressure, and a three-term recursion inequality plus Gronwall's lemma to handle the time-stepping.","core_discovery":"Theorem 3.2: if the saturation constant α of the reduced velocity space Ur and the pressure-gradient space span{∇ψ_1,...,∇ψ_r} is strictly less than 3/8, then the difference between the reduced-order solution and the full finite-element solution satisfies a bound where the summed velocity and pressure errors are at most C(α) e^{((7τ+4α)/2)T} times the sum of the omitted velocity and pressure POD eigenvalues, each weighted by norms of the corresponding POD modes. This makes the POD truncation error the sole driver of the ROM error, up to a stability factor.","pith_inferences":["The saturation constant α is computable from the POD bases; a practical implementation should measure it, because if α ≥ 3/8 the proven bound does not apply and the error may not follow the POD truncation.","The numerical experiments generate POD modes with the L2 inner product, while the theory assumes H1 (gradient) inner products; verifying the theory numerically would require repeating the tests with H1-based POD and checking α.","The exponential factor e^{((7τ+4α)/2)T} implies the error bound is only useful over moderate time horizons; for long-time integration the bound becomes very pessimistic, suggesting the need for time-domain decomposition or a sharper analysis.","The rigorous error analysis is for the linear Stokes problem; the lid-driven cavity test adds a nonlinear convection term via extrapolation, so a full Navier–Stokes error estimate remains open, as the authors themselves indicate."],"forward_implications":["The ROM error can be reduced predictably by adding POD modes; the estimate gives the per-mode contribution as weighted eigenvalues, so mode selection is a matter of truncating small eigenvalues.","No inf-sup/LBB condition is required in the reduced pair; pressure is obtained explicitly through a pressure-stiffness update, preserving the decoupling that makes projection methods efficient.","The scheme is second-order in time; the numerical tables show convergence rates near 1.99 for velocity and 1.98 for pressure.","Very few modes suffice on smooth problems: two velocity modes give relative errors around 10^-7, and about four pressure modes reach the same level.","The method extends to parametric settings: in the lid-driven cavity test, the ROM tracks the projection error when interpolating in Reynolds number and even when extrapolating in time."],"fun_headline_variants":["POD truncation drives Stokes ROM error","α<0.375 keeps reduced Stokes model accurate","Second-order ROM for Stokes with proven error bound","Incremental projection ROM: truncation rules error","Stable projection ROM error tied to POD truncation"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the reduced velocity space and the pressure-gradient space satisfy the saturation property with constant α < 3/8, and that the time step is below a threshold depending on α—conditions neither proven for POD bases nor checked in the numerical experiments, which instead use L2-based POD.","fun_headline_variants_meta":{"raw":{"variants":["POD truncation drives Stokes ROM error","α<0.375 keeps reduced Stokes model accurate","Second-order ROM for Stokes with proven error bound","Incremental projection ROM: truncation rules error","Stable projection ROM error tied to POD truncation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00029,"raw_usage":{"total_tokens":1475,"prompt_tokens":626,"completion_tokens":849,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":370,"completion_tokens_details":{"reasoning_tokens":776}},"tokens_in":370,"tokens_out":849,"duration_ms":8131,"temperature":1.0,"reasoning_tokens":776,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T17:07:52.718933+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute α = sup_{y∈Ur, z∈∇Qr} |(y,z)_X| / (||y||_X ||z||_X) for the H1-based POD spaces built from the paper's snapshot ensembles; if for typical snapshot sets one finds α ≥ 3/8, the theorem's hypothesis is violated and the stated error bound is not guaranteed. Additionally, with α in hand, one can compare the measured ROM errors against the right-hand side of (3.26) at fixed ν and τ to test the bound directly.","supporting_citations":[],"review_version":1}