{"id":"745cfa86-a9ce-4a7f-9794-d71ad6f23276","arxiv_id":"2504.15385","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A kinetic Chapman-Enskog derivation yields two-field hydrodynamic equations for bacterial density and internal CheY-P concentration, predicting nonlocal chemotactic response with a memory-controlled smoothing length.","lead":"Bacteria such as E. coli steer by remembering chemical signals over tens of seconds, so the classic Keller-Segel equations miss key behavior. This paper derives new two-variable equations that capture the resulting nonlocal, memory-driven response and predict a characteristic smoothing length of about 170 to 220 micrometers.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The two-field closure is validated only at τ̂=1, not at the E. coli parameters used for quantitative predictions, where the smoothing length is comparable to the mean free path and the gradient expansion is uncontrolled.","rationale":"The strongest claim is that the two-field equations produce a nonlocal response with a memory-dependent smoothing length, with specific numbers for E. coli. For this to hold, the Chapman-Enskog closure must be accurate at the parameters used. The paper's validation is a single simulation at τ̂=1, λ=1, α1=0, which is far from the E. coli parameter set. The slow-field justification (τ̂≈4.2) is an argument, not a test, and it is weaker than it appears: the first neglected internal mode is only a factor of 2 slower than ρX, and the advection time across the predicted L0 is comparable to the relaxation time of that mode, so the separation of scales is not clean. The internal inconsistency in the reported L0 values (170 vs 220 µm) and the mismatch with the reported transport coefficients support the conclusion that the E. coli numbers are not yet reliable. I considered the alternative concerns: the deferred derivation to Ref. [46] is a verifiability issue, but the Supplemental Material provides the key equations; the traveling-wave comparison is admittedly outside the linear regime and is presented only as a qualitative check. Those are less load-bearing than the closure validity at the prediction parameters. The reader's verdict of CONDITIONAL is appropriate; the condition should be a direct simulation test at E. coli parameters.","tokens_in":17121,"tokens_out":22072,"duration_ms":183421,"concrete_test":"Run agent-based simulations (same algorithm as Fig. 2) with the E. coli parameters τ̂=4.2, λ=1.62, α1=0.33 and a step ligand signal of small amplitude (e.g., l1=0.1), then fit the stationary density profile to the theoretical form ρ(x)=ρ0+ρ0l1 sgn(x)[ψ1+ψ0(1−e^{−k0|x|})] and independently extract L0=1/k0. Also measure the diffusion and mobility coefficients from the same simulations (e.g., via the stationary response to a weak sinusoidal signal) and recompute L0 from Eq. (13); if the fitted L0 differs from the theoretical value by more than about 20%, or if the measured coefficients contradict the reported D's, the two-field closure is not quantitatively valid for the E. coli regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim is the two-field description (Eqs. 7–12) and the nonlocal response length L0 for E. coli. The only direct numerical test of this closure (Fig. 2) runs simulations at τ̂=1, λ=1, α1=0, with L0=0.43 in units of V/ν0, i.e., a Knudsen number of order 2 and no time-scale separation (1/τ equals ν0). The E. coli predictions use τ̂=4.2, λ=1.62, α1=0.33, for which the first neglected internal mode relaxes at 2/τ≈0.105 s⁻¹ (time ~9.5 s), while the advection time across the predicted L0≈170 µm is about 6 s; the slaved mode is not fast relative to the hydrodynamic scale, and ε~V/(ν0L0)≈0.7–1.3 is O(1). The closure is therefore uncontrolled at the parameters of interest, and the Fig. 2 validation does not cover them. Furthermore, the paper reports L0=1.7×10² µm in the numerics section and 2.2×10² µm in the conclusions, and recomputing L0 from the reported D11, D12, D22 and Eq. (13) with γ1=1 gives about 96 µm, so the E. coli numbers are internally inconsistent. If the closure is inaccurate at τ̂=4.2, the predicted smoothing effect, the central new result, is not quantitatively established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This letter proposes a macroscopic description of run-and-tumble bacterial chemotaxis with internal memory. Starting from a kinetic equation for the distribution f(r,n,X,t), with X the normalized CheY-P concentration, the authors apply a Chapman-Enskog expansion and keep two slow fields: the bacterial density ρ and the first CheY-P moment ρX. The resulting hydrodynamic equations (7)-(12) contain cross-diffusion and chemotactic mobility coefficients; in the short-memory limit they reduce to the Keller-Segel model. For the linear model (A=X, B=b, C=e^{λX}), the authors give explicit transport coefficients in the Supplemental Material, a Lorentzian static response function Ψρ(k,0) with smoothing length L0=k0^{-1}, and a nonlocal step-signal density profile. They compare the profile with agent-based simulations at τ̂=1, and compare the predicted traveling-wave current with experiments on E. coli. For E. coli parameters they report numerical values of the transport coefficients and L0≈170-220 µm.","tokens_in":17368,"tokens_out":8388,"duration_ms":68516,"significance":"If the two-field equations are accurate, they provide a macroscopic closure that captures memory-induced nonlocal chemotaxis and they yield testable predictions (step-profile smoothing, nonmonotonic current versus wave speed). The manuscript includes reproducible analytic truncations for the linear model and a direct simulation comparison, which are strengths. However, the central quantitative claim depends on a closure whose validation is performed at parameters different from those used for the E. coli predictions, and the reported smoothing length is inconsistent with the reported transport coefficients. These issues need to be resolved before the result is fully convincing.","major_comments":[{"comment":"The numerical validation of the closure is performed at parameter values far from those used for the quantitative E. coli predictions. Fig. 2 uses τ̂=1.0, λ=1.0, α1=0, and L0=0.43 in units of V/ν0, whereas the E. coli section uses τ̂≈4.2, λ=1.62, α1=0.33. The justification for including ρX as a slow field in the paragraph after Eq. (4) explicitly relies on τ̂≈4.2, so the simulation at τ̂=1 does not test that premise. In addition, with the reported E. coli parameters the dimensionless gradient scale is ε≈V/(ν0 L0) ≈ 0.7-1.3 (depending on whether L0 is taken as 170 µm or 96 µm), so the Chapman-Enskog expansion is not a small-gradient expansion in the regime where the nonlocal response is predicted. The agreement shown in Fig. 2 is encouraging but does not establish the accuracy of the predicted L0 for E. coli.","section":"After Eq. (4); Fig. 2; Numerical values for E. coli"},{"comment":"The value of the smoothing length is internally inconsistent. The Numerical values section states L0=1.7×10^2 µm; the earlier version of the Conclusions states L0=2.2×10^2 µm; the later version of the Conclusions states L0=170 µm. Moreover, substituting the reported coefficients D11=1.3×10^3 µm²/s, D12=0.81×10^3 µm²/s, D22=0.99×10^3 µm²/s, τ=19 s, and γ1=1 into Eq. (13) gives k0 = sqrt(D11γ1/[τ(D11D22−D12^2)]) ≈ 0.0104 µm^{-1}, i.e., L0≈96 µm. This direct computation disagrees with all three reported values. The authors must correct the quoted L0 values and ensure consistency between the transport coefficients and the derived smoothing length.","section":"Numerical values for E. coli; Eq. (13); Conclusions"},{"comment":"The supplementary material gives the linearized equations and the n=1 and n=2 analytical truncations, but the E. coli transport coefficients in the Numerical values section are not accompanied by the truncation order or a convergence test. The figure caption for Fig. 1 mentions truncation at n=10, but the reader cannot verify whether the quoted D and µ values are the n=10 results. Please state the truncation used and provide a small convergence table for the coefficients that enter the reported L0 and traveling-wave prediction.","section":"Supplemental Material; Numerical values for E. coli"}],"minor_comments":[{"comment":"The caption states \"Smoothing length L0 (left) and amplitude ψ0 (right)\", while the surrounding text says \"Amplitude ψ0 (left) and smoothing length L0 (right)\"; the description of the two panels must be made consistent.","section":"Fig. 1 caption"},{"comment":"The manuscript contains duplicated text blocks and duplicated figures (two versions of Fig. 2 and of Fig. 3 with different captions, and repeated \"Analysis\" and \"Numerical values\" sections). These production artifacts should be removed before publication.","section":"Manuscript formatting"},{"comment":"The caption contains a typo \"For thse parameters\" and uses the relation \"⟨X⟩ = ρ/ρX\", which contradicts the definition ⟨X⟩≡ρX/ρ used in the main text.","section":"Fig. 2 caption"},{"comment":"The conclusion contains an unresolved citation \"[?]\" in the sentence \"can be measured for specific systems or computed for other models [?]\"; this placeholder must be replaced.","section":"Conclusions"},{"comment":"Reference [46] is incomplete: it is cited as \"Physical Review E X, X (2024)\" with placeholder volume and page numbers, and must be updated to the published details.","section":"Reference [46]"},{"comment":"In Fig. 2(a) the phrase \"the solid dashed line\" is unclear; presumably the authors mean the solid and dashed lines for the memory-model and Keller-Segel predictions. Also, the Conclusions phrase \"allow to compute\" should be \"allow one to compute\".","section":"Fig. 2(a) and language"}],"recommendation":"major_revision","confidential_remarks":"The paper is essentially a letter summarizing companion Ref. [46]. The internal L0 inconsistency and the missing truncation details are fixable, but the closure validation at off-target parameters is a substantive concern that bears directly on the central quantitative claim. The editors may wish to request the companion paper or its derivation details as part of the review process."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this letter. First, it has a real idea: including the CheY-P moment as a second slow field in a Chapman-Enskog expansion gives two-field hydrodynamic equations whose linear response is nonlocal, with a smoothing length that grows with memory time. That is a genuine extension of Keller-Segel for a regime that matters in soils and marine patches. Second, the quantitative claims for E. coli are not currently trustworthy: the reported smoothing length is internally inconsistent, and the agent-based validation was run at tau_hat=1, not at the E. coli parameters used for predictions.\n\nThe derivation is standard and reasonably transparent. The SM gives explicit Hermite-expansion coefficients for the linear model, so the two-field equations are reproducible in that case. The reduction to KS in the short-memory limit is a good check, and the step-profile simulation at tau_hat=1, lambda=1, alpha1=0 agrees nicely with the theory at that parameter point. The qualitative capture of the nonmonotonic traveling-wave current is suggestive even though the experiment lies outside linear response.\n\nThe problems come when the authors move to E. coli. They quote L0 = 1.7e2 um in the numerics section and 2.2e2 um in the conclusions; recomputing from their own D11, D12, D22 and Eq. (13) with gamma1=1 gives about 96 um. That is a three-way mismatch, not a rounding error. Worse, the only direct test of the closure is at tau_hat=1, where the memory time equals the run time and no time-scale separation exists. At tau_hat=4.2, the first neglected internal mode relaxes in about 9.5 s, the advection time across the predicted L0 is about 6 s, and epsilon ~ V/(nu0 L0) is O(1). The slow-field assumption for rho_X is therefore uncontrolled at exactly the parameter set used for the headline numbers. The authors acknowledge tau_hat=4.2 is large, but they do not quantify the error in the gradient expansion.\n\nNone of this kills the project. The two-field structure is likely right in a qualitative sense, and a full derivation exists in the companion paper. But the letter needs to fix the L0 numbers and either validate at tau_hat=4.2 or present the closure as a controlled approximation only in the short-memory limit. As it stands, I would send it to a serious referee because the idea is worth the referees' time, but I would not cite the E. coli numbers in my own work until they are reconciled.","headline":"A credible two-field extension of chemotaxis with memory, but the E. coli smoothing length is internally inconsistent and the closure is validated only where it is least controlled.","tokens_in":17982,"tokens_out":3324,"would_cite":false,"duration_ms":30655,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that E. coli chemotaxis needs a second macroscopic field, the CheY-P moment, and that memory makes the response nonlocal over $L_0=170$ µm.","keywords":["bacterial chemotaxis","memory effects","run-and-tumble","CheY-P","Chapman-Enskog","nonlocal response","E. coli","smoothing length"],"falsifier":"Measure the steady density profile and local tumbling rate of E. coli in a microfluidic channel with a sharp step in chemoattractant; if the profile cannot be fit by $\\rho_0+\\rho_0 l_1\\,\\mathrm{sgn}(x)[\\psi_1+\\psi_0(1-e^{-k_0|x|})]$ with $L_0\\approx 170\\,\\mu$m, or if a traveling-wave experiment shows no maximum in current versus wave speed near $V_s\\approx 8\\,\\mu$m/s, the two-field memory closure is wrong.","tokens_in":16807,"feed_emoji":"🦠","tokens_out":13987,"duration_ms":107700,"temperature":0.7,"pith_summary":"The paper argues that the standard Keller–Segel chemotaxis equations, which couple bacterial density locally to the chemical gradient, stop being valid once the chemoattractant varies on the tens-of-seconds timescale set by the slow methylation of E. coli chemoreceptors. On that scale, the macroscopic state must include a second field: the first moment of the CheY-P concentration, the protein that controls tumbling. The authors derive coupled reaction-diffusion equations for density and that moment, and the static response to a periodic signal is a Lorentzian in wavenumber, $\\Psi_\\rho(k,0)=\\psi_0/(1+(k/k_0)^2)+\\psi_1$, which becomes a smoothed response in real space with length $L_0=k_0^{-1}$. For measured E. coli parameters this smoothing length is about 170 µm, the same order as pore sizes and nutrient-patch scales in natural habitats. If correct, local chemotaxis models miss a real nonlocal effect in those environments.","feed_headline":"Bacteria remember: chemotaxis is nonlocal over 170 microns","feed_subtitle":"A 19-second memory sets a 170-micron reach that local chemotaxis equations miss.","key_machinery":"The central object is the normalized CheY-P fluctuation $X(t)$, modeled by the Langevin equation $\\dot X=-(A(X,l)+B(X,l)\\dot l)/\\tau+\\sqrt{2/\\tau}\\,\\xi$, with $\\tau\\approx 19$ s the methylation-controlled memory time and $\\dot l$ the ligand change in the swimmer's frame. At the kinetic level the distribution $f(\\mathbf r,\\hat{\\mathbf n},X,t)$ obeys a Fokker–Planck equation in $X$ together with a run-and-tumble collision operator. The load-bearing step is the Chapman–Enskog closure that keeps the first CheY-P moment $\\rho X$ as an approximately slow field because the dimensionless memory time $\\hat\\tau=\\nu_0\\tau\\approx 4.2$ puts the internal relaxation close to the tumbling timescale. This closure produces the two-field equations with transport coefficients expressed as integrals over the kinetic solution; in the linear model the static response function is Lorentzian, $\\Psi_\\rho(k,0)=\\psi_0/(1+(k/k_0)^2)+\\psi_1$, with smoothing length $L_0=k_0^{-1}$.","core_discovery":"On the paper's own terms, the central discovery is that the memory of the internal chemotaxis pathway cannot be removed by a short-gradient expansion; it must be promoted to a hydrodynamic field. Starting from a kinetic equation for the distribution over position, swimming direction, and the normalized CheY-P fluctuation $X$, the authors apply Chapman–Enskog closure with density $\\rho$ and the first CheY-P moment $\\rho X$ as the slow fields. The resulting equations (7)–(12) contain cross-diffusion and cross-mobility terms that vanish as the memory time goes to zero, recovering the classical local model. In the linear regime the static density response is a Lorentzian in wavenumber, implying a characteristic smoothing length $L_0$; a step signal gives the density profile $\\rho(x)=\\rho_0+\\rho_0 l_1\\,\\mathrm{sgn}(x)[\\psi_1+\\psi_0(1-e^{-k_0|x|})]$. The same equations predict that the chemotactic current induced by a traveling wave is non-monotonic in wave speed, with a maximum in the order of the experimentally observed value.","pith_inferences":["Beyond the paper: the Lorentzian response implies a cutoff for short-wavelength density fluctuations, so in a self-consistent model the memory length may suppress the classical chemotaxis aggregation instability; the paper does not discuss this.","Beyond the paper: the same two-field structure should apply to any run-and-tumble organism with a slow adaptation variable, and substituting that organism's $\\tau$, $\\nu_0$, and $\\lambda$ gives a concrete testable prediction for its smoothing length.","Beyond the paper: because $L_0$ is only marginally larger than the run length $V/\\nu_0\\approx 123\\,\\mu$m, a decisive test would measure the step-response profile across mutants with altered methylation rates and check that the fitted $L_0$ grows with memory as shown in Fig. 1."],"forward_implications":["Where the chemoattractant profile varies on scales comparable to $L_0\\approx 170\\,\\mu$m, the standard local model mispredicts both the density profile and the bacterial flux; the two-field equations are the replacement.","A sharp step in ligand concentration produces a density profile that is a smoothed step with exponential tails $e^{-k_0|x|}$, and the local average tumbling rate becomes a measurable proxy for the hidden CheY-P moment.","The response to a traveling chemoattractant wave is non-monotonic in wave speed, with a peak in the same order of magnitude as the measured $V_s\\approx 8\\,\\mu$m/s, unlike the local Keller–Segel prediction.","In the short-memory limit $\\tau\\to0$, the cross-couplings vanish ($D_{12},\\mu_{12}\\to0$) and the classical equations are recovered, so the new description contains the old one as a limit.","The transport coefficients are explicit functions of measurable single-bacterium parameters, so the same equations can be instantiated for other strains or organisms without re-deriving the kinetic closure."],"supporting_citations":[{"why":"Supplies the kinetic equation for $f(\\mathbf r,\\hat{\\mathbf n},X,t)$ and the Langevin dynamics for $X$ that the letter starts from, together with the simulation method used for comparison.","marker":"[8]"},{"why":"Provides the run-and-tumble kinetic coefficients used to compute the diffusion and mobility coefficients in the local-chemotaxis limit.","marker":"[47]"},{"why":"Companion paper containing the full Chapman–Enskog derivation and explicit expressions for the transport coefficients quoted in the letter.","marker":"[46]"},{"why":"Gives the single-cell E. coli parameters $\\nu_0=0.22$ s$^{-1}$, $\\tau=19$ s, $\\lambda=1.62$, and $V=27\\,\\mu$m/s used for the numerical values.","marker":"[6]"},{"why":"Establishes the kinetic-to-local-equation derivation that the present work extends to the memory regime.","marker":"[42]"},{"why":"Reports the measured non-monotonic current versus traveling-wave speed, with a maximum near 8 µm/s, against which the model is compared.","marker":"[57]"},{"why":"Underpins the two-state motor model with tumbling rate $\\nu=\\nu_0 e^{\\lambda X}$ used as the linear model.","marker":"[19]"}],"fun_headline_variants":["Bacteria's 19-second memory makes chemotaxis nonlocal","Chemotaxis equations break for bacteria with memory","Nonlocal chemotaxis: bacteria remember chemical gradients","Keller-Segel invalid for chemotaxis with memory"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The two-field description treats the first CheY-P moment as approximately slow even though the memory time (19 s) and the tumbling reorientation time (about 4.5 s) differ by only a factor of about four; if that time-scale separation is not wide enough, the Chapman–Enskog closure and its transport coefficients are inaccurate.","fun_headline_variants_meta":{"raw":{"variants":["Bacteria's 19-second memory makes chemotaxis nonlocal","Chemotaxis equations break for bacteria with memory","Nonlocal chemotaxis: bacteria remember chemical gradients","Keller-Segel invalid for chemotaxis with memory"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00113,"raw_usage":{"total_tokens":4656,"prompt_tokens":862,"completion_tokens":3794,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":478,"completion_tokens_details":{"reasoning_tokens":3727}},"tokens_in":478,"tokens_out":3794,"duration_ms":23597,"temperature":1.0,"reasoning_tokens":3727,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:28:04.805555+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the steady density profile and local tumbling rate of E. coli in a microfluidic channel with a sharp step in chemoattractant; if the profile cannot be fit by $\\rho_0+\\rho_0 l_1\\,\\mathrm{sgn}(x)[\\psi_1+\\psi_0(1-e^{-k_0|x|})]$ with $L_0\\approx 170\\,\\mu$m, or if a traveling-wave experiment shows no maximum in current versus wave speed near $V_s\\approx 8\\,\\mu$m/s, the two-field memory closure is wrong.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the kinetic equation for $f(\\mathbf r,\\hat{\\mathbf n},X,t)$ and the Langevin dynamics for $X$ that the letter starts from, together with the simulation method used for comparison."},{"cited_title":"Villa-Torrealba, S","cited_arxiv_id":null,"evidence_quote":"Gives the single-cell E. coli parameters $\\nu_0=0.22$ s$^{-1}$, $\\tau=19$ s, $\\lambda=1.62$, and $V=27\\,\\mu$m/s used for the numerical values."}],"review_version":1}