{"id":"d811f336-3e05-4f48-9471-b5a902f19c6f","arxiv_id":"2507.15103","paper_version":2,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A fully discrete mixed finite element method for stochastic Keller-Segel is analyzed, but the claimed O(k^{1/2}+h+k^{-1/2}h^2) rate rests on an invalid expectation step.","lead":"This paper develops a Crank-Nicolson finite element method for a stochastic Keller-Segel chemotaxis model and claims error rates that grow worse when the time step shrinks. The key inverse-time-step effect comes from an expectation calculation that appears to be wrong, so the main claim is not supported.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The h4/k term in Theorem 4.4 rests on an expectation miscalculation: E|ΔWm|2 = k turns the h4|ΔWm|2/k bound into h4, not h4/k.","rationale":"The reader identified exactly the two load-bearing defects, and my independent reading confirms them. The expectation step preceding Eq. (4.14) is the single most important error: a term proportional to |ΔWm|^2/k is bounded in expectation as h^4/k, but E|ΔWm|^2 = k forces the contribution to be h^4. This is not a typographical slip in one constant; it is the only mechanism producing the k^{-1/2}h^2 rate in Theorem 4.4, which the abstract and Section 6 advertise as the main finding. Test 2 is explicitly designed to confirm k^{-1/2} growth with k ≪ h^2, so the numerical experiment is presented as evidence for precisely the term whose derivation is wrong. The second defect, the pathwise vanishing of the Itô integral in Lemma 2.2, is independent and also serious: it underpins the P-a.s. H1 and H2 regularity used in Theorem 3.2 and Lemma 2.3(c). The proof sets an Itô integral to zero 'by the divergence theorem' with no justification at the level of stochastic integrals; even if a hidden integration-by-parts identity held in some distributional sense, it cannot make the martingale term vanish P-a.s. pointwise in time with the coefficient 2δ unfiltered. I do not see a way to preserve the advertised inverse-k dependence, so the verdict should remain REJECT. Credit where due: the deterministic Crank-Nicolson plus splitting mixed FEM construction, the mass-conservation Lemma 4.1, and the stability Lemmas 3.1/4.2 are plausible and would support a weaker O(k^{1/2}+h) strong error result; the numerical experiments are consistent with the corrected rate once Test 2 is reinterpreted as an artifact of comparing Err = C k^{1/2} + C h with fixed h = 0.1 (the observed growth as k shrinks is then the fixed-h floor, not k^{-1/2}). The paper is salvageable as a convergence analysis without the inverse-k term, but not with its central claim intact.","tokens_in":33112,"tokens_out":3046,"duration_ms":27850,"concrete_test":"Re-derive the expectation of the noise splitting term III in the proof of Theorem 4.3 without the hidden replacement E|ΔWm|^2 = k^2. Concretely: keep the bound E[||θ^{m+1/2}||^2 |ΔWm|^2/k] and evaluate it exactly as h^4 E||u^{m+1/2}||^2_{H2}; then re-run the Gronwall summation (4.15) with h^4 in place of h^4/k. If the resulting fully discrete estimate is O(k + h^2) instead of O(k + h^2 + h^4/k), the k^{-1/2}h^2 component of Theorem 4.4 and the interpretation of Test 2 collapse. A supplementary check: re-prove Lemma 2.2(c) retaining, rather than zeroing, the Itô integral; if P-a.s. H1 regularity cannot be closed with the stated constant 7ν/4 − 2δ^2C|b|^2, the regularity basis of the error analysis is also in question.","verdict_should_be":"REJECT","load_bearing_attack":"The advertised central claim is the inverse-k error term O(k1/2 + k−1/2h2), the paper's stated departure from deterministic Crank-Nicolson analysis. The fully discrete analysis locates this term in the noise error estimate for III in Section 4.2, just before Eq. (4.14). The text bounds III by a multiple of h4 ||u^{m+1/2}||^2_{H2} |ΔWm|^2 / k and then, after the words 'which together with Lemma 3.1 implies that', asserts E[III] ≤ C h4/k. But E|ΔWm|^2 = k for a standard Wiener increment, so the expectation of h4 |ΔWm|^2/k is h4/k × k = h4, not h4/k. The displayed line E[III] ≤ (νk/16)E||∇ε^{m+1/2}||^2 + C δ^2|b|^2 K3 h^4/ν is therefore the correct-looking end product of an erroneous intermediate: the h4/k sum contribution in Eq. (4.15) has no valid source. Without that term, Theorem 4.4 and Theorem 4.3 reduce to O(k + h^2); the claimed k^{-1/2}h^2 component, the abstract's 'notably ... inverse dependence on k', and Test 2's interpretation all lose their theoretical basis. A second, independent problem appears in Lemma 2.2: the proof of Part (c) sets the stochastic integral L2 = 2δ ∫ (∇u, ∇(b·∇u)) dW identically to zero 'by the divergence theorem'. This is not justified pathwise; the integrand is not a pathwise divergence and the Itô integral cannot be discarded pointwise in ω. Parts (c)-(d) of Lemma 2.2, used throughout Theorem 3.2 (via C1, C2 and Lemma 2.3(c)), therefore lack a sound proof. Both defects are load-bearing for the headline rate; the first alone suffices to invalidate the advertised result.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a fully discrete Crank-Nicolson splitting mixed finite element method for a stochastic Keller-Segel system with Stratonovich gradient-type multiplicative noise. It claims P-a.s. stability for the semi- and fully discrete schemes and strong error estimates of the form E||u(t_m)-u_h^m||^2 + E[k sum ||grad(u(t_{m+1/2})-u_h^{m+1/2})||^2] <= C(k+h^2+h^4/k), with analogous bounds for the auxiliary variable sigma and the chemoattractant concentration c. The resulting convergence rate is quoted as O(k^{1/2}+h+k^{-1/2}h^2), and numerical experiments are presented for balanced, unbalanced, and small-noise regimes.","tokens_in":33550,"tokens_out":43886,"duration_ms":405083,"significance":"If the results are correct, this would be the first numerical analysis for this particular stochastic Keller-Segel model, and the proof strategy (splitting mixed FEM combined with Crank-Nicolson time stepping and a random projection-error term) is a useful contribution. The paper also provides pathwise stability estimates for the discrete solutions and Monte Carlo validation. Two criticisms that have been raised against the manuscript do not, on close reading, land. First, the expectation step in Eq. (4.14) is not an error: E|Delta W_m|^2 = k makes the factor |Delta W_m|^2/k have expectation 1, so the per-step contribution is O(h^4); the O(h^4/k) appearing in Eq. (4.15) is produced by summing over M = O(k^{-1}) time steps, not by replacing h^4 with h^4/k in a single step. Second, the stochastic integral discarded in Lemma 2.2(c) is genuinely zero pathwise, because for periodic functions and a constant vector b, (grad u, grad(b . grad u)) = (1/2) int_D b . grad(|grad u|^2) dx = 0; the phrase 'by the divergence theorem' is terse but the conclusion is valid.","major_comments":[{"comment":"The coefficient of k sum E||grad e_u^m||^2 changes from (15 nu/32 - delta^2 |b|^2/2) in the per-step inequality (3.42) to (15 nu/64 - delta^2 |b|^2) in the summed inequality (3.43) and in the final Gronwall display (3.44). These coefficients are not equal: summing (3.42) gives (15 nu/32 - delta^2 |b|^2/2) on the diagonal terms together with an additional 11 nu/64 contribution on the shifted index. Moreover, the displayed coefficient (15 nu/64 - delta^2 |b|^2) need not be positive under the theorem's assumption (3.29); it is negative whenever delta^2 |b|^2 > 15 nu/64, which (3.29) permits. Since the Gronwall argument requires a positive energy coefficient, the proof of the O(k) semi-discrete bound (3.30), which feeds directly into Theorem 4.4, is not valid as written. This is fixable by summing the coefficient correctly or by strengthening (3.29), but it is load-bearing.","section":"3.2, Eqs. (3.42)-(3.44)"},{"comment":"In the estimate for epsilon_u^1, the term C k ||sigma_h^{1/2}||^2_{H1} ||epsilon_u^1||^2_{H1} appears on the right-hand side and is then silently replaced by C K1 (K1 + K_tilde) k in the next display. No inequality is supplied to justify this replacement: an inverse estimate would introduce h^{-2}, and absorbing the term into the left-hand side would require an unstated smallness condition on k relative to h and the stability constants. Consequently, the displayed chain does not establish the first-step bound (4.17), which is needed for the fully discrete rate in Theorem 4.3 and hence in Theorem 4.4. This is a genuine gap in the proof as written, although it appears to be fixable with a more careful first-step argument.","section":"4.2, Eq. (4.16)"}],"minor_comments":[{"comment":"The statement that the error estimates are 'sharp' is not supported: no lower bound is proved, and Test 2 compares two discrete solutions that both carry the h^4/k mechanism in their error bounds, so it does not demonstrate that the error relative to the exact solution actually grows like h^4/k.","section":"Introduction and Section 5"},{"comment":"The sentence 'without loss of generality, we assume that u_h^0 = u^0 and sigma_h^0 = sigma^0' is not a valid reduction; the initial conditions of Algorithm 2 are projections. The subsequent proof only needs epsilon_u^0 = 0 and epsilon_sigma^0 = 0, which follow from the definitions, so the sentence should be removed or replaced by the correct statement.","section":"4.2, initial data paragraph"},{"comment":"Part (d) of Lemma 2.2 is left as an exercise for the reader, but it is used in Lemma 2.3(c) and elsewhere; a proof or a precise citation should be supplied.","section":"Lemma 2.2(d)"},{"comment":"Condition (3.4) is described as a condition on u_0 even though it depends on k through kappa_1; it should be identified as a mesh/CFL-type condition involving both the data and the time step.","section":"Lemma 3.1, condition (3.4)"},{"comment":"There are several typographical and wording issues, including 'Prelimimaries' in the Section 2 heading, 'usefull', and 'satisifes'; in addition, the reference solution used in Test 2 is not specified, unlike in Test 1.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the two objections that have been circulated with this manuscript (the h^4/k expectation step and the Ito integral in Lemma 2.2) are, on close reading, not valid objections, and I would not reject the paper on those grounds. The real defects are the coefficient inconsistency in Eq. (3.43) and the uncontrolled H1 term in Eq. (4.16); both appear fixable within the manuscript's framework. The paper would also benefit from toning down the 'sharpness' claim and from clarifying the numerical interpretation of Test 2."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper's distinguishing claim, the O(k^{1/2}+k^{-1/2}h^2) rate with its inverse dependence on k, does not survive a careful read. The bound just before Eq. (4.14) is C h^4 ||u^{m+1/2}||^2 |Delta W_m|^2 / k. Since E|Delta W_m|^2 = k, taking expectations gives h^4, not h^4/k. The displayed h^4/k term in Theorem 4.4 has no valid source. The second load-bearing problem is Lemma 2.2(c): the Ito integral L2 is set to zero 'by the divergence theorem,' but the integrand is not a pathwise divergence and the integral is not zero P-a.s. Parts (c)-(d) of that lemma are used throughout Theorem 3.2, so the semi-discrete analysis is also shaky.\n\nWhat the paper does well: it is, as far as I can tell, the first numerical analysis for this stochastic Keller-Segel system. The Crank-Nicolson plus splitting mixed FEM setup is reasonable, the stability estimates and mass conservation are coherent, and the experiments are run carefully. The numerical rates in Tests 1 and 3 are consistent with what one would expect after the correction, but Test 2 specifically interprets the artifact (errors growing as k shrinks) as evidence for the inverse-k term. That interpretation loses its theoretical basis once Eq. (4.14) is fixed.\n\nThe soft spots are in proportion: the central advertised result is wrong as stated, and the proof error is not a typo—it is a sign that the h^4/k dependence was imported into the bound without checking the Wiener increment moment. If the inverse-k term is removed, the error estimates become O(k+h^2), which is still a useful, if less flashy, contribution. There are also several skipped proofs in the discrete stability section, but those are routine and not the main issue.\n\nThe citation pattern looks normal: the external well-posedness results [14,17] are cited appropriately, and the splitting-mixed technique is attributed to prior work. There is self-citation, but it is not doing any questionable work here.\n\nWho this is for: researchers working on numerical methods for stochastic PDEs, especially chemotaxis models, might find the scheme and the corrected-rate version valuable. But the paper as is should not be cited for the inverse-k result. I would send it to peer review—the errors are substantive and a referee can confirm them—but the author needs to be told, plainly, that Theorem 4.4 must be corrected and the advertised rate revised before publication.","headline":"The advertised inverse-k rate in Theorem 4.4 rests on an expectation miscalculation at Eq. (4.14), and Lemma 2.2 sets an Ito integral to zero pathwise; both gaps are load-bearing.","tokens_in":34039,"tokens_out":1950,"would_cite":false,"duration_ms":25110,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N12","65N15","65N30"],"pacs":[],"model":"deepseek-v4-flash","headline":"A fully discrete Crank-Nicolson finite element method for the stochastic Keller-Segel system is proven to converge strongly at rate $O(k^{1/2}+h+k^{-1/2}h^2)$.","keywords":["stochastic Keller-Segel","chemotaxis","gradient-type multiplicative noise","Stratonovich noise","Crank-Nicolson method","splitting mixed finite element","strong error estimates","finite-time blow-up"],"falsifier":"Compute the expectation of the noise error term in equation (4.14) directly: since $\\mathbb{E}|\\Delta W_m|^2 = k$, the term $h^4 \\|u^{m+1/2}\\|^2 |\\Delta W_m|^2 / k$ has expectation $h^4 \\mathbb{E}\\|u^{m+1/2}\\|^2$, not $h^4/k$; if this is the correct bound, then the $h^4/k$ term and the $k^{-1/2}$ factor in Theorem 4.4 disappear. Alternatively, run the scheme with fixed $h$ and successively smaller $k$: if the errors do not grow as $k$ decreases, the predicted inverse-$k$ behavior is absent.","tokens_in":32908,"feed_emoji":"🦠","tokens_out":9861,"duration_ms":97356,"temperature":0.7,"pith_summary":"This paper develops the first numerical analysis for a stochastic Keller-Segel chemotaxis system with gradient-type multiplicative Stratonovich noise, proposing a fully discrete scheme that combines the Crank-Nicolson time discretization with a splitting mixed finite element method. The main claim is a strong convergence rate of order $O(k^{1/2} + h + k^{-1/2}h^2)$ for the cell density, the auxiliary gradient variable, and the chemical concentration, where the $k^{-1/2}$ term is a genuinely stochastic effect absent in deterministic chemotaxis analysis. The paper proves stability, mass conservation, and identifies that first-order convergence is achieved when the time step satisfies $k = h^2$, with error growth when $k$ is much smaller than $h^2$. Numerical experiments support the predicted rates and also illustrate finite-time blow-up under large initial data. A sympathetic reader would regard this as the first rigorous convergence framework for numerical simulation of this stochastic chemotaxis model.","feed_headline":"Stochastic chemotaxis solver's error grows when time step shrinks","feed_subtitle":"Proof gives a noise-induced inverse time-step term, so balancing k=h^2 is needed for first-order accuracy.","key_machinery":"The argument is carried by a splitting mixed finite element formulation in which the auxiliary variable $\\sigma = \\nabla c$ is introduced, rewriting the elliptic equation as a mixed system and removing the need for an LBB inf-sup condition; linear finite element spaces are then chosen freely. Time discretization uses the Crank-Nicolson midpoint rule, the natural approximation for Stratonovich noise. The error analysis is organized around an error equation for the projected (finite element) components, with three load-bearing pieces: $L^2$ projections $P_u$, $P_\\sigma$, $P_c$ that convert the continuous error into a finite-dimensional one with $O(h^2)$ interpolation error; a nonlinear discrete Gronwall inequality (Lemma 2.1) that controls the non-Lipschitz, sign-indefinite chemotaxis term; and a bound on the noise error term that produces the inverse-$k$ factor $h^4/k$.","core_discovery":"The central result is Theorem 4.4, which bounds the fully discrete strong error by $\\max_{1\\le m\\le M} \\mathbb{E}\\|u(t_m)-u_h^m\\|^2_{L^2} + \\mathbb{E}\\big[k \\sum_m \\|\\nabla(u(t_{m+1/2})-u_h^{m+1/2})\\|^2\\big] \\le C(k+h^2+h^4/k)$, with analogous estimates for $\\sigma = \\nabla c$ and $c$. This yields the strong convergence rate $O(k^{1/2} + h + k^{-1/2}h^2)$. The distinctive $h^4/k$ contribution arises from the stochastic forcing: in the error equation, the noise term is bounded by a multiple of $h^4 \\|u^{m+1/2}\\|^2 |\\Delta W_m|^2 / k$, and after taking expectations this is claimed to give $h^4/k$. The theorem therefore predicts that the method converges at first order when $k = h^2$, and that refining the time step beyond this balance degrades the error estimate, a prediction the paper's Test 2 explicitly seeks to confirm.","pith_inferences":["If the expectation step behind the $h^4/k$ term is corrected, the true rate would likely become $O(k^{1/2}+h)$, eliminating the $k^{-1/2}$ penalty and making $k=h^2$ unnecessary; this is an inference beyond the paper's stated claim.","The splitting-mixed Crank-Nicolson framework may transfer to other stochastic chemotaxis models or to stochastic convection-diffusion systems with gradient-type noise, since the machinery for the non-Lipschitz nonlinearity is not specific to Keller-Segel.","The pathwise regularity lemma sets an It\\^o integral to zero via the divergence theorem; if that step is not valid pathwise, the regularity assumptions would need re-establishment through martingale arguments, which would leave the numerical scheme unchanged but re-ground the analysis.","Test 4's observation that noise accelerates blow-up is a qualitative phenomenon that could be quantified, for instance by estimating the expected blow-up time as a function of noise intensity $\\delta$."],"forward_implications":["When $k = h^2$, the scheme achieves first-order strong convergence $O(h)$ for $u$, $\\sigma$, and $c$, providing a balanced choice of temporal and spatial steps.","For $k \\ll h^2$, the error bound grows like $k^{-1/2}$, so refining time alone does not improve accuracy and can degrade it, a marked departure from deterministic chemotaxis solvers.","The fully discrete method conserves total mass and operates without an LBB inf-sup condition, making its finite element spaces easy to construct.","At small noise intensity, the numerical experiments recover deterministic-like accuracy: second order for $u$ and $c$ in $L^2$ and first order for $\\sigma$ in $H^1$."],"supporting_citations":[{"why":"Supplies existence, uniqueness, and pathwise regularity of solutions to the stochastic Keller-Segel system, giving the solution framework the error analysis builds on.","marker":"[14]"},{"why":"Provides mass conservation, positivity, blow-up threshold, and the $L^p$ bound used in Lemma 2.2.","marker":"[17]"},{"why":"Introduces the splitting mixed finite element idea for Keller-Segel models that the paper adapts.","marker":"[27]"},{"why":"Supplies the mixed finite element framework and the equivalent $H^1$ norms for the auxiliary variable without an LBB condition.","marker":"[5]"},{"why":"Gives the nonlinear discrete Gronwall lemma (Lemma 2.1) underpinning all stability estimates.","marker":"[19]"},{"why":"Provides the reference framework for fully discrete mixed finite element error analysis with gradient-type multiplicative noise and the comparison method used in numerical tests.","marker":"[6]"}],"fun_headline_variants":["Shrinking time step worsens chemotaxis solver error bound","Noise term makes smaller time steps inflate chemotaxis error","Stochastic chemotaxis method: error rises as k falls","In stochastic chemotaxis, h^4/k term punishes tiny k"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the noise error term has expected size $h^4/k$ rather than $h^4$, which requires treating $|\\Delta W_m|^2$ in the bound as an $O(1)$ quantity even though $\\mathbb{E}|\\Delta W_m|^2 = k$.","fun_headline_variants_meta":{"raw":{"variants":["Shrinking time step worsens chemotaxis solver error bound","Noise term makes smaller time steps inflate chemotaxis error","Stochastic chemotaxis method: error rises as k falls","In stochastic chemotaxis, h^4/k term punishes tiny k"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000876,"raw_usage":{"total_tokens":3792,"prompt_tokens":950,"completion_tokens":2842,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":566,"completion_tokens_details":{"reasoning_tokens":2766}},"tokens_in":566,"tokens_out":2842,"duration_ms":23559,"temperature":1.0,"reasoning_tokens":2766,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:41:41.193460+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the expectation of the noise error term in equation (4.14) directly: since $\\mathbb{E}|\\Delta W_m|^2 = k$, the term $h^4 \\|u^{m+1/2}\\|^2 |\\Delta W_m|^2 / k$ has expectation $h^4 \\mathbb{E}\\|u^{m+1/2}\\|^2$, not $h^4/k$; if this is the correct bound, then the $h^4/k$ term and the $k^{-1/2}$ factor in Theorem 4.4 disappear. Alternatively, run the scheme with fixed $h$ and successively smaller $k$: if the errors do not grow as $k$ decreases, the predicted inverse-$k$ behavior is absent.","supporting_citations":[{"cited_title":"Huang and J","cited_arxiv_id":null,"evidence_quote":"Supplies existence, uniqueness, and pathwise regularity of solutions to the stochastic Keller-Segel system, giving the solution framework the error analysis builds on."},{"cited_title":"Misiats, O","cited_arxiv_id":null,"evidence_quote":"Provides mass conservation, positivity, blow-up threshold, and the $L^p$ bound used in Lemma 2.2."},{"cited_title":"Zhang, J","cited_arxiv_id":null,"evidence_quote":"Introduces the splitting mixed finite element idea for Keller-Segel models that the paper adapts."},{"cited_title":"Duarte-Rodr ´ ıguez, M.´A","cited_arxiv_id":null,"evidence_quote":"Supplies the mixed finite element framework and the equivalent $H^1$ norms for the auxiliary variable without an LBB condition."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the nonlinear discrete Gronwall lemma (Lemma 2.1) underpinning all stability estimates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the reference framework for fully discrete mixed finite element error analysis with gradient-type multiplicative noise and the comparison method used in numerical tests."}],"review_version":1}