{"id":"010b3a6c-de8c-40be-9a83-c2339895475a","arxiv_id":"1908.05600","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper derives the mean, variance, PDF, and CDF of the time-variant channel impulse response for a 3D diffusive molecular communication link with two mobile nodes and an absorbing receiver, and applies them to drug delivery and communication design.","lead":"This paper derives formulas for the random channel between two moving nanomachines that communicate by releasing molecules, where the receiver absorbs the molecules. The formulas are then used to optimize drug release schedules and communication settings for mobile molecular systems.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The free-diffusion distance PDF (8) is inherited by every theorem; its hard-core violation is only validated for r0 ≫ arx + atx, leaving the central stochastic characterization unverified where Tx starts near Rx.","rationale":"The reader's weakest_assumption identifies the same load-bearing point: equation (8), and hence all derived statistics, uses a free-diffusion distance distribution that violates the geometric exclusion between Tx and Rx. My stress-test confirms this is the most load-bearing concern because it enters every claimed result — the mean, variance, PDF, and CDF of h(t,τ) and the PDF and CDF of p(t,Tb) — whereas the other flagged issue (Lemma 4's monotonicity proof) affects only one design step in Section V-D and does not threaten the central stochastic characterization. I do not treat the approximation as a fatal flaw: the paper explicitly acknowledges it in Remark 2, and the particle-based simulation with reflection shows good agreement for the paper's chosen parameters, which is genuine independent support for that parameter regime. The concern is one of scope: the simulation validates r0 = 10 μm with arx + atx = 1.1 μm, where the forbidden mass is minuscule, but the abstract and theorems make a general claim. The concrete test above would determine whether the formulas degrade when the initial distance is comparable to the contact radius, a regime that is physically plausible for targeted drug delivery where carriers are injected close to diseased cells. If the test shows negligible bias, the conditional acceptance is fully justified; if it shows significant bias, the paper should add an explicit validity condition on r0 and D2t relative to arx + atx. I therefore recommend keeping the reader's CONDITIONAL verdict unchanged: the central derivation is sound where validated, but the scope of the claim should be conditioned on the parameter regime until the hard-core approximation is tested there.","tokens_in":23378,"tokens_out":9844,"duration_ms":105154,"concrete_test":"Run the paper's particle-based simulation (reflecting Tx) for r0 = 2 μm, arx = 1 μm, atx = 0.1 μm, DRx = 0, DTx = 1e-14 m^2/s, and DX = 8e-11 m^2/s, at t = 36 s and τ = 0.17 s (or a few (t,τ) pairs), and compare the empirical mean and PDF of h with Theorem 1 and Theorem 2. Also compute M(t) = ∫_0^{arx+atx} frptq(r)dr from (8). If M(t) is non-negligible (say > 1%) and the analytical mean deviates from simulation by more than the Monte Carlo error, the central formulas need a stated validity regime; if M(t) stays below 1% and agreement holds, the approximation concern is settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central stochastic characterization (Theorems 1–2 and Corollaries 1–3) all inherit the free-space relative-distance PDF frptq(r) in (8). This PDF has support r ≥ 0, whereas the physical model has an absorbing receiver and a transparent transmitter of positive radius, so the true distance PDF must vanish for r < arx + atx and is distorted near contact by reflection (as [25] and Remark 2 concede). Remark 2 attempts to dismiss this by noting that (8) tends to 0 as r → 0; this only controls the density at the origin, not the probability mass ∫_0^{arx+atx} frptq(r)dr nor the distortion of the boundary layer just above arx + atx. The bias is not uniform in h: for small τ, h(τ|r) is sharply peaked at r ≈ arx, exactly the region where the hard-core constraint is most active. Since h is monotone decreasing for r beyond the peak, even a few percent of forbidden mass maps to the upper tail of f_h and to the CDF of p, where the paper's design evaluations are most sensitive. The numerical validation (Figs. 2–3) uses r0 = 10 μm, arx = 1 μm, atx = 0.1 μm; at the plotted t values the forbidden mass is negligible, so those simulations do not exercise the fragile regime r0 ≈ arx + atx or D2t ≈ r0^2. Thus the conditions under which (8) is a good approximation are narrower than the unqualified central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes a 3D diffusive mobile molecular communication (MC) system in which a spherical transparent transmitter, a spherical absorbing receiver, and signaling molecules all undergo Brownian motion. The authors derive the mean, variance, PDF, and CDF of the time-variant channel impulse response h(t,tau), and the PDF and CDF of the probability p(t,Tb) that a released molecule is absorbed during a bit interval. These stochastic channel results are then applied to two design problems: a controlled-release drug delivery system that minimizes the total released drug amount subject to a target absorption rate, and an MC system with imperfect CSI where the detection threshold, release profile, and bit-frame duration are optimized. The analytical results are compared with particle-based and Monte Carlo simulations for the chosen system parameters.","tokens_in":23684,"tokens_out":17477,"duration_ms":179351,"significance":"The paper's main derivations in Section III are direct probabilistic computations from the noncentral chi-distributed Tx-Rx distance and the known absorbing-sphere CIR formula, with no fitted parameters; this is a strength and makes the mean, variance, PDF, and CDF of h and p potentially useful building blocks for mobile MC design. The authors also honestly acknowledge in Remark 2 that the free-diffusion distance PDF is an approximation because it ignores steric exclusion of the transmitter by the receiver. If the application-level flaws identified below are corrected, the framework would be a useful contribution to mobile molecular communication and drug-delivery modeling. In its current form, however, the paper contains several nontrivial errors in the application theorems that undermine the claimed design guarantees.","major_comments":[{"comment":"The distance PDF f_r(r) in Eq. (8) is the free-diffusion noncentral chi density, which assigns positive probability to r < atx + arx even though the physical model has a hard-core exclusion between the transmitter and receiver. Remark 2 dismisses this by noting that (8) tends to zero as r tends to zero, but that does not control the integrated forbidden mass or the distortion of the boundary layer near r = atx + arx. Since the CIR in Eq. (1) is sharply peaked near r = arx for small tau, this is exactly the region where the approximation error matters most. All subsequent results (Theorems 1-2, Corollaries 1-3) inherit this bias. The numerical validation in Figs. 2-3 uses r0 = 10 um, arx = 1 um, atx = 0.1 um, where the forbidden mass is negligible at the plotted times, so it does not exercise the fragile regime r0 close to arx + atx or D2 t comparable to r0^2. The paper should either state and justify the parameter regime in which (8) is accurate, or quantify the approximation error.","section":"Section III-A, Eq. (8), Remark 2"},{"comment":"The convolution formula (29) for the PDF of g(t) assumes that the terms alpha_i h(t_i, t - t_i) are independent across release times. Under the stated model, h(t_i, t - t_i) depends on r(t_i), and the random variables r(t_i) for different i are correlated because the transceivers follow continuous Brownian motion. Thus g(t) is a sum of dependent random variables and Theorem 3 is not valid for the model described in Section II. The drug-delivery performance evaluation in Fig. 6 relies on this theorem, so the corresponding analytical claims are not established. The Monte Carlo description in Section VI-A is also ambiguous: it must be clarified whether different release times are sampled from a single Brownian trajectory or independently, since only the latter would agree with the convolution formula but would be inconsistent with the model.","section":"Section III-D, Theorem 3, Eq. (29)"},{"comment":"The proof of Lemma 4 claims that dF_{r(t)}(r)/dt < 0 for all r. This is not true in general: for a fixed r < r0, the CDF F_{r(t)}(r) starts at 0 and initially increases as diffusion brings probability mass below r. Even for r > r0, the derivative expression in (44) can be positive when the term -2n - 3/2 + (x^2 + r0^2)/(4D2 t) becomes positive for large t and small n. Therefore the proof is incorrect, and the monotonicity used to obtain the maximum bit-frame duration in Section V-D is not justified unless additional restrictions on r and t are provided.","section":"Appendix E, Lemma 4"},{"comment":"The convexity claim in Lemma 3 is not correct as stated. With a = p(1-p), b = eta, c = xi - eta, one computes the second derivative of zeta(alpha) = (c - p alpha)/sqrt(2a alpha + 2b) and obtains zeta''(alpha) = [-5a^2 p alpha - 8abp - 3a^2 c]/(2a alpha + 2b)^{5/2} < 0, so zeta is concave. For erf(zeta) to be convex, the term -2 zeta (zeta')^2 in its second derivative must dominate zeta'', and this is not guaranteed; for example, with p = 0.5, eta = 1, xi = 2, the second derivative of erf(zeta(alpha)) is negative at alpha = 10. Consequently, the objective in (38) is not established to be convex, and the claim that the interior-point method yields a global optimum is unsupported.","section":"Appendix D, Lemma 3"}],"minor_comments":[{"comment":"The CDF in Eq. (9) is printed with sqrt(2 DTx t) in the Marcum Q-function, whereas Eq. (5) and the proof in Appendix A, Eq. (43), use sqrt(2 D2 t). This matters when DRx > 0, which is the case in the MC application in Section VI-C; please correct the typo.","section":"Lemma 1, Eq. (9)"},{"comment":"The paragraph describing the Monte Carlo setup after Fig. 3 should specify whether r(t_i) at different release times are generated from correlated Brownian paths or independently. The current wording is ambiguous and is directly relevant to whether the simulation validates Theorem 3 or the independent-release assumption.","section":"Section VI-A"},{"comment":"The 'second moment' expression in Eq. (14) is an unevaluated integral involving 1/r1, which is acknowledged in Remark 4. Since this integral is a central object in the variance computation, please provide numerical integration details or bounds on the truncation error.","section":"Corollary 1, Remark 4"},{"comment":"The PDF formula in (17) uses the notation h'(r, tau) and h_hat without a clear formal definition of h_hat as the deterministic mapping in Eq. (1). A short sentence defining h_hat(r, tau) explicitly would improve readability.","section":"Theorem 2, Eq. (17)"}],"recommendation":"major_revision","confidential_remarks":"The most serious issue is the independence assumption in Theorem 3, which is a fundamental modeling error for Brownian motion and directly invalidates the drug-delivery performance analysis. The convexity claim in Lemma 3 and the monotonicity claim in Lemma 4 are also incorrect as written. If the authors cannot replace these with valid arguments, the paper should be rejected. I recommend a major revision only if the authors can correct or remove the invalid application theorems while preserving the core channel statistics."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper gives the first stochastic characterization of the time-variant CIR for a 3D diffusive mobile MC system with an absorbing receiver: mean, variance, PDF, and CDF of the CIR, plus PDF/CDF of the per-period absorption probability. That is a real building block for the mobile MC subfield, and the two applications—optimized release profiles for drug delivery and OOK threshold/release/frame-length design under imperfect CSI—are concrete and show meaningful gains. The derivations are honest probabilistic computations from the noncentral chi distance distribution and the known absorbing-sphere CIR, with no fitted parameters, and simulation matches analysis for the presented settings.\n\nThe main soft spot is that every formula inherits the free-space distance PDF (8), which lets the transmitter diffuse through the receiver. Remark 2 acknowledges this but the justification is too quick: the density at r→0 going to zero does not make the probability mass within the excluded volume negligible. For small τ the CIR is sharply peaked near r = arx, precisely where the hard-core constraint binds. The numerical validation uses r0 = 10 μm against arx + atx = 1.1 μm and short times, so it does not exercise the fragile regime. The abstract's unqualified claim should be tempered with the condition that matters (e.g., r0 well above the radius sum, or D2 t not too large), or the error should be bounded.\n\nLemma 4's proof has a gap: the series term contains (-2n-3/2 + (x^2+r0^2)/(4D2t)), which is not always non-positive, so the claimed global monotonicity of Frptq(~r(ψ)) in t does not follow from the written argument. The plotted parameters satisfy it, but the proof should be fixed or the condition stated. The variance is also left as a numerical integral; that is fine, but the abstract should not oversell it as fully closed-form.\n\nNo code or data are shipped, which is a minor reproducibility minus. The self-citations to [12] and [25] are legitimate; the new piece is the distributional result, not the building blocks. All told, the central claim holds for the regime simulated, and the paper is a solid contribution to the MC community. I would send it to peer review with a request to tighten the approximation statement and repair Lemma 4, then accept.","headline":"First full stochastic characterization of the time-variant CIR for 3D mobile MC with absorbing receivers—useful, mostly sound, but the hard-core approximation and Lemma 4 proof need attention.","tokens_in":24234,"tokens_out":3808,"would_cite":true,"duration_ms":37173,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper derives the mean, variance, PDF, and CDF of a mobile molecular channel's time-variant impulse response with an absorbing receiver, then applies them to drug delivery and OOK design under imperfect CSI.","keywords":["molecular communication","diffusive mobile channel","absorbing receiver","channel impulse response","stochastic analysis","drug delivery","on-off keying","time-variant channel"],"falsifier":"Run a particle-based simulation with a reflecting boundary at the receiver surface (the transmitter bounces back on collision) using an initial distance $r_0$ only slightly larger than $a_{\\mathrm{Tx}}+a_{\\mathrm{Rx}}$ and a large $D_{\\mathrm{Tx}}$, so that collisions are frequent, and compare the empirical mean and PDF of $h(t,\\tau)$ with Theorem 1 and Theorem 2; a systematic deviation that grows as $r_0$ approaches $a_{\\mathrm{Tx}}+a_{\\mathrm{Rx}}$ would show the free-diffusion approximation is inadequate in close-encounter regimes.","tokens_in":23168,"feed_emoji":"🧪","tokens_out":12058,"duration_ms":109235,"temperature":0.7,"pith_summary":"This paper establishes that a 3D diffusive mobile molecular communication channel with a freely diffusing transmitter, an absorbing receiver, and freely diffusing signaling molecules has a time-variant impulse response whose mean, variance, PDF, and CDF can be derived in closed or semi-closed form. The key step is treating the Tx-Rx distance as a noncentral-chi random variable and pushing its distribution through the deterministic absorbing-receiver hitting-rate function. This yields, for any release time, the full distribution of the channel impulse response and of the probability that a released molecule is absorbed within a given interval, without Monte Carlo averaging. The paper uses these statistics to minimize the number of drug molecules released while keeping the expected absorption rate minus a safety margin above a target, and to optimize detection threshold, release profile, and bit-frame duration for on-off-keying with outdated channel state information. Numerically, the optimized designs save 27–54% of the drug mass relative to constant release and cut the maximum bit error rate by up to a factor of 8.","feed_headline":"Full statistics derived for mobile molecular channels","feed_subtitle":"Closed-form mean, PDF, and CDF enable optimized drug release and on-off-keying system design.","key_machinery":"The load-bearing object is the random Tx-Rx distance $r(t)$, whose PDF is the free-diffusion noncentral-chi (generalized Rayleigh) density $$f_{r(t)}(r)=\\frac{r}{r_0\\sqrt{\\pi D_2 t}}\\exp\\left(-\\frac{$r^{2}$+$r_0^{2}$}{4D_2 t}\\right)\\$\\sinh$\\left(\\frac{r_0 r}{2D_2 t}\\right),$$ with $D_2 = D_{\\mathrm{Tx}}+D_{\\mathrm{Rx}}$ the effective relative diffusion coefficient. Around this object the paper builds a change-of-variables mechanism: the deterministic CIR $$h(t,\\tau)=\\frac{a_{\\mathrm{rx}}}{\\sqrt{4\\pi D_1\\$tau^{3}$}}\\left(1-\\frac{a_{\\mathrm{rx}}}{r}\\right)\\exp\\left(-\\frac{(r-a_{\\mathrm{rx}})^2}{4D_1\\tau}\\right)$$ is unimodal in $r$, so the PDF and CDF of $h$ are obtained by evaluating the distance PDF and CDF at the two inverse distances $r_1(h)<r_2(h)$ and summing. For the absorption probability $p(t,T_b)$, the same map is monotone decreasing in $r$, so only one inverse branch is needed. This transformation is what converts known single-particle hitting-rate formulas into a full stochastic channel characterization.","core_discovery":"The central discovery is that the randomness of all three diffusing entities collapses into a single random variable, the distance $r(t)$ between the transmitter and receiver centers, which follows a noncentral chi distribution with three degrees of freedom, Eq. (8). Because the absorbing-receiver CIR $h(t,\\tau)$ is a deterministic, unimodal function of $r(t)$ — it rises as the receiver approaches the optimal distance and falls as it moves away — the statistics of $h$ follow from a two-branch transformation: every sub-maximum value $h$ corresponds to two distances, one on each side of the distance that maximizes the CIR. The mean of the CIR is obtained in closed form (Theorem 1), the variance via a numerically integrable second-moment expression (Corollary 1), and the PDF and CDF as explicit compositions of the distance PDF and CDF with the two inverse branches (Theorem 2 and Corollary 2). The same machinery, applied to the monotone relation between distance and absorption probability $p(t,T_b)$, yields the PDF and CDF of $p$ (Corollary 3), and these feed the drug-delivery and communication optimization problems.","pith_inferences":["The two-branch transformation is not specific to the CIR: any channel quantity that is unimodal in the Tx-Rx distance, such as per-molecule mutual information, could be given a PDF and CDF by the same inverse-branch construction, with monotone quantities such as $p(t,T_b)$ as the one-branch special case.","The Chebyshev-based bound $P_\\theta(t)\\ge 1-1/\\beta^2$ is conservative by design; a designer who instead evaluates the derived CDF of $g(t)$ can choose $\\beta$ from the actual tail of the absorption rate and potentially release fewer molecules for the same reliability.","Because the distance statistics depend on the diffusion coefficients only through $D_2=D_{\\mathrm{Tx}}+D_{\\mathrm{Rx}}$ and the CIR depends on them through $D_1=D_X+D_{\\mathrm{Rx}}$, an experiment that varies receiver mobility while holding these sums fixed would test whether the derived channel statistics are invariant under that trade-off; this symmetry is not explicitly discussed in the paper.","The model treats the receiver as a perfectly absorbing sphere; replacing it with a reversible-adsorption surface would break the closed-form CIR expression, but the inverse-branch machinery would survive as long as a monotone or unimodal distance-to-surface statistic can be identified."],"forward_implications":["With only the initial Tx-Rx distance and the diffusion coefficients known, the full time-dependent distribution of the received absorption rate becomes computable, so release schedules for drug delivery can be designed offline with no real-time position tracking.","The optimized drug release profile is tri-phasic — large initial release, then reduction, then increase as the carrier diffuses away — and uses 27–54% fewer molecules than constant release while meeting the target absorption rate.","Designs that ignore transmitter mobility fail to keep the expected absorption rate above the target for most of the treatment window, so mobility must be included for reliable drug delivery.","In OOK threshold detection with imperfect CSI, optimizing the detection threshold and release profile lowers the maximum frame BER by up to a factor of 8 relative to uniform release for $A=10^5$ molecules and $T=3000$ s.","The maximum bit-frame duration that keeps molecule usage efficiency above a target can be found from the CDF of the distance distribution, since the probability that a released molecule is absorbed above a threshold decreases with release time."],"supporting_citations":[{"why":"Supplies the deterministic absorbing-receiver CIR $h(t,\\tau)$ and absorption probability $p(t,T_b)$ formulas whose statistics the paper derives.","marker":"[2]"},{"why":"Establishes the time-variant stochastic channel modeling framework for diffusive mobile MC that this work extends to absorbing receivers.","marker":"[12]"},{"why":"Provides the effective diffusion coefficients $D_1$ and $D_2$ and the reflecting-boundary distance PDF against which the free-diffusion approximation (8) is checked.","marker":"[25]"},{"why":"Gives the generalized Rayleigh (noncentral chi) distribution used to obtain the Tx-Rx distance statistics in Lemma 1.","marker":"[28]"},{"why":"Provides the definite integrals used to evaluate the mean CIR in closed form in Theorem 1.","marker":"[31]"},{"why":"Supplies the convexity rules used to prove that the optimization problems for detection threshold and release profile are convex.","marker":"[32]"},{"why":"Gives the function-of-random-variable PDF formula and the Chebyshev inequality used in the CIR/absorption-probability distributions and the performance bound.","marker":"[34]"}],"fun_headline_variants":["Mobility collapses to one random variable in MC analysis","Closed-form channel stats for diffusive mobile MC","Distance-based statistics optimize molecular drug delivery","Single distance variable drives molecular channel randomness"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything downstream inherits the free-diffusion distance PDF (8), which lets the transmitter diffuse through the receiver instead of being reflected at its surface; if close transmitter-receiver encounters are frequent, the mean, variance, PDF, and CDF in Theorems 1-2 and Corollaries 1-3 become biased.","fun_headline_variants_meta":{"raw":{"variants":["Mobility collapses to one random variable in MC analysis","Closed-form channel stats for diffusive mobile MC","Distance-based statistics optimize molecular drug delivery","Single distance variable drives molecular channel randomness"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000624,"raw_usage":{"total_tokens":2945,"prompt_tokens":1059,"completion_tokens":1886,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":675,"completion_tokens_details":{"reasoning_tokens":1839}},"tokens_in":675,"tokens_out":1886,"duration_ms":15188,"temperature":1.0,"reasoning_tokens":1839,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:22:27.152848+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a particle-based simulation with a reflecting boundary at the receiver surface (the transmitter bounces back on collision) using an initial distance $r_0$ only slightly larger than $a_{\\mathrm{Tx}}+a_{\\mathrm{Rx}}$ and a large $D_{\\mathrm{Tx}}$, so that collisions are frequent, and compare the empirical mean and PDF of $h(t,\\tau)$ with Theorem 1 and Theorem 2; a systematic deviation that grows as $r_0$ approaches $a_{\\mathrm{Tx}}+a_{\\mathrm{Rx}}$ would show the free-diffusion approximation is inadequate in close-encounter regimes.","supporting_citations":[{"cited_title":"Three-dimensional channel characteristics for molecular communications with an absorbing receiver,","cited_arxiv_id":null,"evidence_quote":"Supplies the deterministic absorbing-receiver CIR $h(t,\\tau)$ and absorption probability $p(t,T_b)$ formulas whose statistics the paper derives."},{"cited_title":"Stochastic channel modeling for diffusive mobile molecular communication systems,","cited_arxiv_id":null,"evidence_quote":"Establishes the time-variant stochastic channel modeling framework for diffusive mobile MC that this work extends to absorbing receivers."},{"cited_title":"Diffusive mobile molecular communications over time-variant channels,","cited_arxiv_id":null,"evidence_quote":"Provides the effective diffusion coefficients $D_1$ and $D_2$ and the reflecting-boundary distance PDF against which the free-diffusion approximation (8) is checked."},{"cited_title":"Generalized Rayleigh processes,","cited_arxiv_id":null,"evidence_quote":"Gives the generalized Rayleigh (noncentral chi) distribution used to obtain the Tx-Rx distance statistics in Lemma 1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the definite integrals used to evaluate the mean CIR in closed form in Theorem 1."},{"cited_title":"Boyd and L","cited_arxiv_id":null,"evidence_quote":"Supplies the convexity rules used to prove that the optimization problems for detection threshold and release profile are convex."},{"cited_title":"Papoulis and S","cited_arxiv_id":null,"evidence_quote":"Gives the function-of-random-variable PDF formula and the Chebyshev inequality used in the CIR/absorption-probability distributions and the performance bound."}],"review_version":1}