{"id":"1bf00244-c3f5-4859-b491-06797b9a1c17","arxiv_id":"1908.03876","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A formal adjoint-based multiple-relaxation-time lattice Boltzmann framework is derived for optimal control of coupled chemotaxis systems, but the claimed recovery of the governing equations holds only under an unverified negligible-term assumption and the discrete gradient is not shown to be the…","lead":"This paper derives mathematical formulas for steering chemotaxis models, where cells, bacteria, or criminals move in response to chemical or environmental cues, using the fast lattice Boltzmann simulation method. It provides adjoint equations and gradient formulas for optimal control, but it contains no computer experiments and leaves a key approximation term unquantified.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 1's Chapman-Enskog recovery is conditional on a never-estimated deviation term; for x,t-dependent convection tensors in the paper's own examples the term is generically nonzero and alters the recovered equation.","rationale":"The decisive issue is internal to the proof of Proposition 1. Recovery of (1) is the prerequisite for both the primal MRT solver and the adjoint/AMRT gradient derivation, so a gap here is load-bearing. The proof derives Eq. (63) and then introduces DV; instead of estimating it, the proof states that if it is null or negligible then (63) reduces to (64). This is an explicitly conditional result, while Proposition 1 is stated unconditionally. For the general class in (1), T_Theta is allowed to depend on x,t, and the paper's own examples (2.2.2-2.2.4) use a spatially varying flow field omega. With T_Theta=a Theta e_1, C_Theta determined by (37), DV=beta Theta e_1, and the spurious term is of the same tau-order as the diffusion term. Unless a precise tau(epsilon) scaling is supplied and shown to suppress this term, the model does not recover (1). I therefore agree with the reader's first weakest assumption. The absence of numerical experiments is a separate weakness, but the DV gap is sufficient to reject the central claim. The verdict stays REJECT, so no adjustment to the reader's verdict is needed.","tokens_in":37329,"tokens_out":7387,"duration_ms":81685,"concrete_test":"On D2Q9 take T_Theta(x,t;Theta)=a(x,t)Theta e_1 with a(x,t)=1+beta t, and fix d_Theta and K_Theta so that D_Theta=d_Theta C_s^2 tau (K_Theta^{-1}-1/2 I) is the target tensor. Compute C_Theta=a^2 Theta e_1 e_1^T from (37), substitute into (60) to obtain DV=beta Theta e_1, and write out Eq. (63). If the extra term -tau div_1((K_Theta^{-1}-1/2 I)beta Theta e_1) is nonzero, the recovered equation contains a spurious anisotropic diffusion term that is not in (1) and is of the same order in tau as the target diffusion. Repeat with a spatially varying a(x,t), e.g. a=1+alpha x, to confirm that the x,t-dependence of T_Theta, not the particular beta choice, is the source. If the authors believe a specific tau(epsilon) scaling makes this term O(epsilon^3), they must state and prove that scaling as part of the test.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 1 is the foundation of the paper: if (30)-(32) do not recover (1), the MRT primal solver and the entire AMRT/adjoint workflow have no target PDE. The proof's Eq. (63) leaves an unremoved term -tau div_1((K_Theta^{-1}-1/2 I)DV(x,t;Theta)), with DV defined in (60) as dT_Theta/dt_1 + (div_1(C_Theta) - div_1(T_Theta)dT_Theta/dTheta). Nothing in the hypotheses on T_Theta, D_Theta, C_Theta forces DV=0; the paper merely states 'If this term is null or negligible...'. For the admissible class, T_Theta may depend explicitly on x,t (e.g. the flow velocity omega in Examples 2.2.2-2.2.4), and C_Theta is determined by the integrability condition (37), which does not control DV. In a concrete D2Q9 case with T_Theta=a(x,t)Theta e_1 and a=1+beta t, one gets C_Theta=a^2 Theta e_1 e_1^T and DV=beta Theta e_1, so Eq. (63) contains a spurious diffusion-like term -tau div_1((K_Theta^{-1}-1/2 I)beta Theta e_1) of the same order in tau as the target diffusion term. Since no scaling relation tau(epsilon) or bound on DV is supplied, the claimed recovery to O(epsilon^3) is not established for the stated problem class.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a continuous and discrete adjoint-based optimal control framework for a class of nonlinear coupled anisotropic convection-diffusion chemotaxis-type systems (NCACDCS). The author derives first-order necessary optimality conditions using sensitivity and adjoint calculus, then proposes a multiple-relaxation-time lattice Boltzmann method (MRT) for the primal problem and an adjoint MRT model (AMRT) for the adjoint problem. The central claims are that, through a Chapman-Enskog expansion, the MRT scheme recovers the target macroscopic system to order O(ε^3), and that the AMRT formulation provides a discrete gradient of the cost functional suitable for gradient-descent optimization. The paper also sketches several application examples from biology, criminology, and tumor modeling.","tokens_in":37677,"tokens_out":5615,"duration_ms":60154,"significance":"If the derivations were sound, the paper would supply a useful unified MRT-LBM/AMRT workflow for optimal control of a broad class of chemotaxis-type systems, extending the author's earlier work in [7] and the MRT literature. The paper is self-contained in setting up the continuous optimality system and gives explicit formulas for the equilibrium distributions and adjoint equilibrium distributions; these are useful and appear formally coherent. The application examples are relevant and cover a wide range of intended uses. However, the central recovery statement in Proposition 1 and the discrete adjoint gradient formula in Theorem 3 are not established as written, and the manuscript contains no numerical verification of the proposed algorithms. The intended contribution is therefore not realized in the present form.","major_comments":[{"comment":"The proof of Proposition 1 is incomplete at the step from Eq. (63) to Eq. (64). The term -τ div_1((K_Θ^{-1}-1/2 I)DV(x,t;Θ)) is discarded based only on the assertion that it is 'null or negligible', but no hypothesis in the proposition and no estimate in the proof implies this. For T_Θ depending explicitly on x and t, as in Examples 2.2.2-2.2.4 where the flow velocity ω enters the convection terms, DV in Eq. (60) is generically nonzero. For instance, in a D2Q9 setting with T_Θ = a(x,t)Θ e_1 and a = 1 + βt, Eq. (37) gives C_Θ = a^2 Θ e_1 e_1^T and DV = βΘ e_1, so the dropped term is a spurious τ-order term of the same formal size as the target diffusion term. Since no scaling relation between τ and ε is provided, this residual cannot be absorbed into the claimed O(ε^3) error, and recovery of the target system (1) from Eqs. (30)-(32) is not established for the general class stated.","section":"Section 3.1, Proposition 1 (Eqs. (60), (63), (64))"},{"comment":"The discrete gradient formula (89) is presented as the gradient of the discretized cost functional Jh in Eq. (88), but the adjoint system in Theorem 3 is derived by varying the continuous LBE (65) and integrating by parts in time and space; it is not obtained by differentiating the fully discrete scheme (66). In particular, the discrete primal (66) contains the force correction (τ^2/2)(∂/∂t + κ_x e_i·∇)F_{Θ,i}, which is absent from the continuous LBE (65) on which the adjoint derivation is based. The backward discrete equation described after Eq. (86) is therefore not the true discrete adjoint of (66), and no consistency or error analysis is supplied for the mismatch. The paper must either derive the exact adjoint of the discrete scheme (66) or prove that the continuous adjoint gradient agrees with the discrete gradient to a controlled order; as written, the 'discrete adjoint' claim is unsupported.","section":"Section 3.2, Theorem 3 and Eq. (89)"},{"comment":"The manuscript contains no numerical experiments for either the MRT primal solver or the AMRT optimization loop, despite the abstract and conclusion asserting that the method is 'reliable, efficient, practical to implement' and despite the algorithmic description in Section 3.2.1. In a numerical-methods paper, such claims require at least a validation test on a model problem, a convergence study, or a comparison with a standard discretization; without one, the practical value of the proposed workflow is unsubstantiated.","section":"Sections 3.1-3.2 and Section 4"}],"minor_comments":[{"comment":"In Eq. (20), the second and third integrals on the right-hand side multiply ∂Φ_v/∂f_2 and ∂Φ_w/∂f_3 by \tilde u; they should multiply by \tilde v and \tilde w, respectively.","section":"Section 2.3, Eq. (20)"},{"comment":"The text refers to 'Fig. 1' for the collision-streaming process, but no figure appears in the manuscript; the figure should be included or the reference removed.","section":"Section 3.1"},{"comment":"In Remark 7, 'bellows' should read 'below'.","section":"Section 2.4, Remark 7"},{"comment":"The CFL-type condition τ ≤ C h^2 is stated without derivation or numerical evidence; as written it is an assertion rather than a justified stability condition.","section":"Section 3.1, Remark 12"},{"comment":"The notation is confusing because \tilde u^* denotes both the full adjoint vector (\tilde u^*, \tilde v^*, \tilde w^*) and its first scalar component in the boundary conditions of the displayed adjoint system; a different symbol for the vector would improve readability.","section":"Section 2.3, Theorem 2"}],"recommendation":"reject","confidential_remarks":"The paper's central new claims are not independently supported: the Chapman-Enskog recovery is conditional on an unestimated deviation term that is nonzero for the paper's own examples, and the 'discrete' adjoint gradient is derived from the continuous LBE rather than from the fully discrete scheme. The absence of any numerical validation further prevents the claims of reliability and efficiency from being assessed. These are load-bearing issues that would require substantial new analysis and experimentation to resolve, so I recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [colleague],\n\nThe reader's report is largely right: the central consistency claim in Proposition 1 has a real hole, and the adjoint derivation stops short of a true discrete adjoint. But this is not a throwaway paper. There is solid formal work here.\n\nWhat is actually new: the paper extends the author's earlier AMRT framework and existing MRT convection-diffusion models to a three-equation anisotropic cross-diffusion chemotaxis system, with controls in the source terms. The continuous adjoint analysis (Theorem 2) is careful and complete, and the construction of the equilibrium distribution with the integrability condition (37) for anisotropic convection is a systematic piece of calculus. If you want a worked example of how to derive gradient formulas for LBM discretizations of coupled PDE systems, the derivation is instructive.\n\nThe soft spots are exactly where the reader puts them. In Proposition 1 the proof defines DV at equation (60) and then says: if this term is null... or is negligible, (63) becomes (64). For TΘ depending on x or t, which the examples allow (the flow velocity in (7)-(9)), DV is generically nonzero and sits at the same order as the target diffusion term. The stress-test example with TΘ=(1+βt)Θe1 gives DV=βΘe1, so the recovered PDE picks up a spurious diffusion-like term. No estimate, no scaling relation, no subclass restriction is given. Proposition 1 as stated—O(ε^3) recovery for the general class—is not established, and this is load-bearing because the MRT primal solver's whole purpose is to reproduce (1).\n\nThe adjoint issue is real but less severe: the gradient (70) comes from the continuous LBE (65), while the actual primal solve uses the discretization (66) with the τ²/2 force term. So (89) is an approximate discrete gradient, not the exact derivative of the discrete cost (88). The paper never quantifies the mismatch, but uses the phrase 'discrete adjoint' more strongly than the derivation supports.\n\nThere are no numerical experiments. 'Reliable, efficient, practical' in the abstract is a promise, not a demonstration.\n\nNet: this deserves a serious referee—the machinery is coherent and the gaps are specific and fixable. As submitted I would not accept it: the Chapman-Enskog result needs a controlled estimate on DV or an explicit restriction to TΘ independent of x,t, the adjoint needs a genuine discrete-adjoint consistency check, and a 2D test case would show whether the method actually works. That is a major-revision agenda, not a rejection of the idea.","headline":"A systematic formal extension of adjoint MRT-LBM to chemotaxis optimal control, but the Chapman-Enskog recovery is not established for the general class and no numerics back the practical claims.","tokens_in":38182,"tokens_out":4701,"would_cite":false,"duration_ms":50040,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92B05","49J20","35Q92","49M25","92C50","65K15","35K57","65M99"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that optimal control of coupled anisotropic chemotaxis systems can be solved with a multiple-relaxation-time lattice Boltzmann method and its adjoint, recovering the continuous equations and supplying a discrete gradient…","keywords":["optimal control","chemotaxis","multiple-relaxation-time lattice Boltzmann method","adjoint-based optimization","Chapman-Enskog expansion","anisotropic convection-diffusion","sensitivity analysis","gradient descent"],"falsifier":"Compute the $DV$ operator from (60) for a manufactured problem with $T_\\Theta = t\\,(x_1,x_2)$ on the D2Q9 lattice and measure $\\tau\\,\\mathrm{div}\\bigl((K_\\Theta^{-1}-\\tfrac12 I)DV\\bigr)$ relative to the retained terms; if it exceeds $O(\\varepsilon^3)$, the LBM solutions cannot converge to (1). A second check: compare the discrete adjoint gradient (89) with a finite-difference gradient of $J_h$ in (88) on the same test case; any mismatch that does not shrink with the discretization contradicts the asserted exactness.","tokens_in":37062,"feed_emoji":"🦠","tokens_out":12603,"duration_ms":119074,"temperature":0.7,"pith_summary":"This paper tries to establish that a broad class of optimal control problems governed by coupled, anisotropic chemotaxis-type systems can be solved numerically by a multiple-relaxation-time (MRT) lattice Boltzmann method paired with an adjoint MRT model. The payoff would be a gradient-descent workflow in which each optimization iteration needs one forward lattice-Boltzmann solve and one backward adjoint solve, with the same local, parallel-friendly structure as the primal scheme. The paper derives first-order optimality conditions for a general three-equation system, builds an MRT scheme whose Chapman-Enskog expansion is claimed to recover the macroscopic equations, and then derives an adjoint MRT scheme whose discrete gradient is claimed to be exact for the discretized cost functional. Applications are drawn from cell migration, tumor invasion, and crime-pattern models.","feed_headline":"One backward lattice Boltzmann pass gives chemotaxis control gradients","feed_subtitle":"Each step needs one forward solve and one backward solve, reusing the same lattice machinery.","key_machinery":"The carrying object is the D2Q9 (two-dimensional, nine-velocity) multiple-relaxation-time lattice Boltzmann scheme with the collision matrix $\\Pi_\\Theta = \\mathrm{diag}(\\Lambda_{\\Theta,1},\\Lambda_{\\Theta,2},\\Lambda_{\\Theta,3})$ in moment space; the block $\\Lambda_{\\Theta,2}$ containing the parameters $k_{\\Theta,ij}$ is what turns the scheme from isotropic into anisotropic diffusion. The adjoint multiple-relaxation-time (AMRT) model is the same type of system run backward in time with the adjoint equilibrium $\\psi^\\mathrm{eq}_{\\Theta,i} = \\sum_k ((\\hat\\Lambda_\\Theta^*)^{-1} O_\\Theta \\hat\\Lambda_\\Theta^*)_{ik}\\psi_{\\Theta,k}$. The Chapman-Enskog expansion of the distribution function in the Knudsen number $\\varepsilon$ is what links the microscopic streaming-and-collision steps to the macroscopic system (1), and the same expansion supplies the choices $\\vec V_\\Theta$, $A_{\\Theta,k}$, and $K_\\Theta$ that produce the target diffusion and cross-diffusion tensors.","core_discovery":"On the paper's own terms, the central discovery is that the discrete MRT lattice Boltzmann equations (30)-(32) approximate the continuous chemotaxis-type system (1) through the Chapman-Enskog expansion with an error of order $O(\\varepsilon^3)$, provided the convection tensors are chosen so that the extra term $\\tau\\,\\mathrm{div}\\bigl((K_\\Theta^{-1}-\\tfrac12 I)DV\\bigr)$ is null or negligible. Given that recovery, the paper defines an adjoint multiple-relaxation-time lattice Boltzmann model in which adjoint distributions stream backward in time along $-\\mathbf{e}_i$, the adjoint equilibrium is built from the transpose of the collision operator, and the gradient of the cost functional is expressed directly in terms of the adjoint distribution functions via (89). This yields a discrete gradient that the paper asserts is exact for the discretized functional, and hence a complete adjoint-based optimization loop for the primal system.","pith_inferences":["A reader could test the recovery claim by evaluating $\\tau\\,\\mathrm{div}\\bigl((K_\\Theta^{-1}-\\tfrac12 I)DV\\bigr)$ for a time- and space-dependent convection tensor; the paper does not estimate this term, so the range of validity of Proposition 1 is an open question, not a proven fact.","The $\\tau^2/2$ force term in the discrete primal scheme (66) is absent from the continuous LBE (65) from which the adjoint is derived; deriving a fully discrete AMRT directly from (66) would either confirm or correct the exactness of gradient formula (89).","For applications where the convection tensors are independent of space and time, such as leading-order crime-pattern models, the problematic $DV$ term may vanish, so the method's practical range may be cleaner than the general theorem's statement.","Coupling the scheme with proper orthogonal decomposition, which the paper mentions as a possibility, is a natural next step that could reduce the memory and CPU cost of the backward-in-time adjoint solves."],"forward_implications":["Each gradient iteration costs one forward MRT solve and one backward AMRT solve; the gradient formula (89) is assembled locally from distribution functions, preserving the parallel structure of lattice Boltzmann methods.","The framework extends to systems with $N\\geq 4$ equations and to mixed or Robin boundary conditions by changing the boundary treatment, so the same code can be adapted across applications.","The discrete adjoint gradient enables gradient and conjugate-gradient descent with line search on the discretized cost functional, including the quadratic case where the step size can be computed explicitly.","For the crime-model, attraction-repulsion, two-species, and tumor-invasion examples, the method gives an implementable route to estimating source terms and parameters from observations."],"supporting_citations":[{"why":"Supplies the earlier adjoint-based MRT formulation for a nonlinear coupled anisotropic convection-diffusion system that this paper extends to chemotaxis-type systems.","marker":"[7]"},{"why":"Supplies the coupled lattice Boltzmann approach for the Keller-Segel chemotaxis model that the primal MRT scheme builds on.","marker":"[81]"},{"why":"Supplies the MRT anisotropic diffusion framework, including the block relaxation matrix and the way anisotropic tensors enter the collision process.","marker":"[84]"},{"why":"Supplies the MRT model for general nonlinear anisotropic convection-diffusion equations, used here for the force and cross-diffusion terms and the Chapman-Enskog derivation.","marker":"[88]"},{"why":"Supplies the moment transformation matrix and the theoretical basis for the dispersion and stability of the MRT scheme used in the D2Q9 lattice.","marker":"[48]"},{"why":"Supplies the adjoint calculus and first-order necessary conditions (the variational inequality and gradient formulas) that the paper applies before discretization.","marker":"[13]"}],"fun_headline_variants":["Backward pass yields chemotaxis control gradients","Adjoint lattice Boltzmann for optimal chemotaxis control","One forward, one backward: control gradients via MRT","Reuse the lattice: adjoint pass for chemotaxis control","Gradient from adjoint lattice Boltzmann for chemotaxis"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The recovery proof of Proposition 1 drops the term $\\tau\\,\\mathrm{div}\\bigl((K_\\Theta^{-1}-\\tfrac12 I)DV\\bigr)$ as null or negligible although it is generically nonzero for space- and time-dependent convection, and the discrete gradient (89) is derived from the continuous LBE (65) rather than from the discrete scheme (66), which contains an extra $\\tau^2/2$ force term.","fun_headline_variants_meta":{"raw":{"variants":["Backward pass yields chemotaxis control gradients","Adjoint lattice Boltzmann for optimal chemotaxis control","One forward, one backward: control gradients via MRT","Reuse the lattice: adjoint pass for chemotaxis control","Gradient from adjoint lattice Boltzmann for chemotaxis"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000533,"raw_usage":{"total_tokens":2609,"prompt_tokens":1033,"completion_tokens":1576,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":649,"completion_tokens_details":{"reasoning_tokens":1496}},"tokens_in":649,"tokens_out":1576,"duration_ms":11356,"temperature":1.0,"reasoning_tokens":1496,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:59:31.258878+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the $DV$ operator from (60) for a manufactured problem with $T_\\Theta = t\\,(x_1,x_2)$ on the D2Q9 lattice and measure $\\tau\\,\\mathrm{div}\\bigl((K_\\Theta^{-1}-\\tfrac12 I)DV\\bigr)$ relative to the retained terms; if it exceeds $O(\\varepsilon^3)$, the LBM solutions cannot converge to (1). A second check: compare the discrete adjoint gradient (89) with a finite-difference gradient of $J_h$ in (88) on the same test case; any mismatch that does not shrink with the discretization contradicts the asserted exactness.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the coupled lattice Boltzmann approach for the Keller-Segel chemotaxis model that the primal MRT scheme builds on."},{"cited_title":"Yoshida, M","cited_arxiv_id":null,"evidence_quote":"Supplies the MRT anisotropic diffusion framework, including the block relaxation matrix and the way anisotropic tensors enter the collision process."},{"cited_title":"Zhenhua, S","cited_arxiv_id":null,"evidence_quote":"Supplies the MRT model for general nonlinear anisotropic convection-diffusion equations, used here for the force and cross-diffusion terms and the Chapman-Enskog derivation."},{"cited_title":"Lallemand, L.S","cited_arxiv_id":null,"evidence_quote":"Supplies the moment transformation matrix and the theoretical basis for the dispersion and stability of the MRT scheme used in the D2Q9 lattice."},{"cited_title":"Belmiloudi, Stabilization, optimal and robust control","cited_arxiv_id":null,"evidence_quote":"Supplies the adjoint calculus and first-order necessary conditions (the variational inequality and gradient formulas) that the paper applies before discretization."}],"review_version":1}