{"id":"197abde4-70ea-4484-b7c5-1dd9b4b7c042","arxiv_id":"2509.02869","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A distributed Voronoi-based gradient algorithm with a decaying step size and threshold-based movement improves probabilistic (Elfes-model) coverage in hybrid wireless sensor networks, in two simulations without a convergence proof.","lead":"This paper proposes a distributed algorithm that moves mobile sensors in a wireless network to improve coverage, using a sensing model where detection probability fades with distance. Each sensor decides its next position using only neighbor information and its local coverage gradient, and two simulations show better coverage than sensors that never move.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Gradient formula (15) omits the boundary contribution from moving Voronoi edges ∂Π'_i, so Algorithm 1 is not a true gradient ascent on the local coverage F_i.","rationale":"I read the paper's central claim as asserting that Algorithm 1 performs distributed gradient ascent on a local coverage objective F_i, with the global coverage F non-decreasing as a result. The Reader's weakest assumption identified the exactness of the coverage decomposition (13) in Lemma 2, and I agree that this is a serious gap. However, I found a more fundamental problem: even if (13) were true, Theorem 1's gradient formula (15) is incomplete because it drops the boundary term from the moving Voronoi edges ∂Π'_i. This is visible from the paper's own setup: the boundary of Π'_i ∩ D(s_i) is listed as including edges of Π_i, but the final formula only integrates over ∂D(s_i) and obstacle-generated segments. A finite-difference check on a two-sensor no-obstacle configuration would settle whether (15) is the true gradient. Since the paper can in principle be repaired by adding the missing boundary term, and the simulations may still indicate heuristic value, the existing CONDITIONAL verdict remains appropriate. I therefore do not change the verdict, but the condition should explicitly include correcting or justifying Theorem 1, not merely the Lemma 2 decomposition.","tokens_in":11062,"tokens_out":8579,"duration_ms":104491,"concrete_test":"Use one mobile sensor s_i at x=(0,0), one stationary sensor s_j at (1,0), no obstacles, φ=1, Elfes r_min=0, r_max=2, α=1. Numerically integrate F_i = ∫_{Π'_i∩D(s_i)} φ p_i(q) dq at x and at x+(δ,0) with δ=1e-4 to estimate the true directional derivative. Compute the x-component of the right-hand side of (15)–(16) at x using numerical integration. If the two differ by more than 1% of |∇F_i|, the missing ∂Π'_i boundary term is confirmed. Repeat with r_max large enough that the Voronoi edge lies strictly inside D(s_i) to ensure the term is active.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing gap is in Theorem 1's gradient derivation. By the Leibniz rule, ∇_{x_i} F_i must include boundary integrals over ∂(Π'_i ∩ D(s_i)), and that boundary includes the moving Voronoi edges ∂Π'_i. The text before Theorem 1 explicitly lists 'edges of Π_i' as part of the boundary, but formulas (15)–(16) contain no integral over those edges; only ∂D(s_i) and obstacle-vertex cuts appear. This is not a minor omission: unlike the global-sum case, where boundary terms cancel between adjacent Voronoi cells, F_i is a single cell integral, so the shifting bisector between mobile sensor i and a neighbor contributes a nonzero boundary term. Consequently (15) is not the true gradient of F_i, and Algorithm 1 is not actually gradient ascent on the local objective. This breaks the chain from local improvement to global monotonicity in Remark 3 even if the Lemma 2 decomposition (13) were exact. The concern is independent of the disjointness issue raised by the Reader; it is a correctness gap in the central mechanism itself.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a distributed deployment algorithm for a hybrid wireless sensor network in which sensing follows the Elfes probabilistic model and obstacles block line of sight. The authors partition the ROI using Voronoi concepts, define a local coverage integral F_i over a locally assigned region, compute a gradient formula (15)-(16), and move each mobile sensor along that gradient with a dynamically scheduled step size, accepting a move only if F_i increases by at least epsilon. They claim the global coverage F can be decomposed into local terms, that local optimization of F_i maximizes F, and that Algorithm 1 converges to a steady-state local-maximum configuration. Two simulations show improved area and weighted coverage relative to the initial static deployment.","tokens_in":11366,"tokens_out":7912,"duration_ms":96727,"significance":"If the theoretical claims were sound, the paper would provide a practical, distributed algorithm for a realistic coverage problem with probabilistic sensing and obstacles. The two simulation examples are encouraging as a proof of concept, and the step-size schedule with an acceptance threshold is a sensible engineering choice. The paper also builds on the authors' prior work, citing the relevant sources. However, the correctness of the central mechanism is not established: the coverage decomposition and the gradient derivation contain gaps that affect the main optimality and convergence claims. The paper is not yet ready for publication in its current form.","major_comments":[{"comment":"Lemma 1 is false in the presence of obstacles. The proof uses only Euclidean distance monotonicity and ignores the visibility constraint Phi(x_i) in Eq. (2). A point q can be closer to x_i than to all other sensors while x_i is occluded from q; then psi_i(q)=0, but a farther visible sensor may have psi_j(q)>0, so p_s(q) != psi_i(q). Consequently, Eq. (10) and the stationary term in Eq. (13) do not represent the true coverage F.","section":"Section III, Lemma 1 and Eq. (10)"},{"comment":"The assertion that the mobile local regions Pi'_i ∩ D(s_i) are disjoint and jointly exhaust the mobile contribution is not proved and is false without an explicit assumption linking r_c to the sensing radii. If r_c < 2 r_s^max, two sensors can have overlapping D(s_i) without being communication neighbors; then a point in the overlap can be assigned to both Pi'_i and Pi'_j. Similarly, a stationary sensor outside the mobile sensor's neighbor set can dominate a point, so the stationary and mobile terms can double-count. The proof's 'by construction' sentence is not a derivation. Add an explicit assumption (e.g., r_c >= 2 r_s^max) and prove disjointness, or revise the decomposition.","section":"Section III, Lemma 2 and Eq. (13)"},{"comment":"The claimed gradient is incomplete. Applying the Leibniz rule to F_i = ∫_{Pi'_i ∩ D(s_i)} φ psi_i dq yields a boundary integral over ∂(Pi'_i ∩ D(s_i)), which includes the moving Voronoi edges ∂Pi'_i inside D(s_i). No such integral appears in (16a)-(16c); the boundary components listed before Theorem 1 explicitly mention edges of Pi_i, but the formula drops them. Unlike in the global coverage sum, these single-cell boundary terms do not cancel. Therefore (15) is not ∇_{x_i} F_i, and Algorithm 1 is not a gradient ascent on F_i. This breaks the local-maximum interpretation of the stopping condition.","section":"Section IV, Theorem 1, Eqs. (15)-(16)"},{"comment":"The claim that F is non-decreasing and converges in finite time depends on Lemma 2 and on the candidate direction being an ascent direction for the true local objective. Since Lemma 2 is unproved and the gradient is incomplete, a move that increases the surrogate F_i need not increase F. The finite-time convergence argument therefore lacks a valid Lyapunov function. Either restore an exact decomposition and an exact gradient, or reframe the algorithm as a heuristic and remove the convergence claims.","section":"Section IV, Remark 3"}],"minor_comments":[{"comment":"'V oronoi' should be 'Voronoi' (spacing error). Also, 'complexifies' in Section I is informal.","section":"Throughout"},{"comment":"The norm in the step-size formula is typeset incorrectly ('|∇xi Fi∥' should be '∥∇xi Fi∥'), and the expression η0 t e^{-βt} ∥∇xi Fi∥ looks dimensionally odd: it mixes the gradient norm with a time-dependent factor. Clarify the intended units and check the formula.","section":"Section IV, Eq. (18)"},{"comment":"The statement 'the boundary of this region is assumed to comprise...' is an assumption, not a derivation. Please justify why these are the only boundary segments, especially in view of the moving Voronoi edges.","section":"Section IV, before Theorem 1"},{"comment":"The simulation parameters are chosen by trial and error, and no comparison with existing algorithms or statistical variation over multiple random initializations is provided. A sensitivity analysis or at least error bars over several runs would make the empirical claims more robust.","section":"Section V"}],"recommendation":"major_revision","confidential_remarks":"The paper's theoretical core relies heavily on the authors' prior work [10], [12] for boundary-term arguments, but those arguments are not made self-contained here. The novelty is mainly the combination of the Elfes model, obstacles, and a dynamic step size; the editor may wish to weigh whether the incremental combination is sufficient for the venue. The simulation study is limited to two configurations and lacks comparisons, so the empirical claims should also be moderated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this is a plausible engineering paper with a real gap in the theory and a misleading gradient derivation. The new pieces are the Elfes-model gradient term, the dynamic step size, and the threshold acceptance rule, wrapped around the Voronoi-based distributed deployment framework from the authors' earlier work. The Elfes gradient calculation is correct, and the simulations show substantial coverage improvements over the static start: weighted coverage from 15% to 48% and area coverage from 27% to 90% in the first example. That is real evidence the algorithm does something useful in practice.\n\nThe main theoretical problem is that formula (15) is not actually the gradient of the local coverage F_i. The Leibniz rule boundary term includes a contribution from the moving Voronoi edges of Π'_i, and the text lists those edges as part of the boundary, but formulas (16a)-(16c) contain no integral over them. The proof waves at prior work that covers only the sensing-range and obstacle terms. Unlike the global-sum case, where boundary terms cancel between cells, the local objective F_i has no such cancellation, so the omitted term is genuinely nonzero. This breaks the stated connection from local gradient ascent to global monotonicity in Remark 3. It's a correctness gap in the central mechanism, not a minor omission.\n\nThere are also two other soft spots. Lemma 1's proof ignores the visibility constraint in (2); when the nearest sensor is occluded by an obstacle, ps(q) is not psi(q), so the decomposition (10) overcounts or undercounts coverage. And Lemma 2 asserts without proof that the mobile local regions Π'_i ∩ D(s_i) are disjoint, which can fail when sensing ranges overlap between non-neighbor sensors. Both are load-bearing for the claimed equivalence between maximizing F and maximizing the sum of local F_i.\n\nThe simulations are decent but only compare against the static initial deployment. No comparison to virtual force or the authors' earlier deterministic gradient methods, so we don't know what the Elfes term and dynamic step size actually buy. Parameters are hand-tuned by trial and error, which is normal for this kind of work but worth stating.\n\nOverall, the paper is a reasonable incremental advance that deserves a serious referee, not a desk reject. The simulations are honest and the practical scheme may work, but the theoretical claims need either a corrected gradient derivation or a softened statement that this is a heuristic with a validated acceptance rule that keeps F non-decreasing. I'd send it to peer review, ask for those changes, and see if the authors can tighten Lemma 2. If they can't, the paper should be reframed as a heuristic study.","headline":"A plausible engineering extension with a real gap in the gradient derivation and unproven convergence claims; the simulations show useful coverage gains, but the theory needs tightening before publication.","tokens_in":11823,"tokens_out":4654,"would_cite":false,"duration_ms":50968,"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":"A distributed gradient rule lets mobile sensors raise coverage using only local information.","keywords":["mobile sensor deployment","probabilistic sensing","Elfes model","Voronoi partition","gradient-based optimization","coverage control","obstacle visibility","hybrid wireless sensor network"],"falsifier":"Take a small network with two mobile sensors whose sensing ranges overlap but whose communication radius is smaller than the distance between them, so they are not neighbors, plus one stationary sensor. Compute the true global coverage F directly by integrating φ(q)·max_i psi(q), then compute the right-hand side of equation (13); any positive discrepancy shows the decomposition and therefore the gradient's target undercounts coverage. Alternatively, place one obstacle so that the closest sensor to a point q is occluded while a farther sensor has line of sight: at that point ps(q) > psi(q) for","tokens_in":10925,"feed_emoji":"📡","tokens_out":5519,"duration_ms":60915,"temperature":0.7,"pith_summary":"The paper addresses a practical question in wireless sensor networks: when sensors have probabilistic, distance-decaying detection and obstacles block sensing, how can mobile sensors reposition to improve overall coverage without a central coordinator? The authors claim that by partitioning the region into local Voronoi cells, the global coverage objective can be split into local terms, and each mobile sensor can ascend the gradient of its own local term using only information from nearby sensors. Their rule moves a sensor only when the predicted local gain exceeds a threshold, uses a step size that first grows then decays, and projects illegal positions back to the visible feasible region. If correct, this gives a scalable, energy-conscious, obstacle-aware deployment method that improves on static placement and converges to a configuration where no sensor and its neighbors can improve by moving.","feed_headline":"Gradient rule lifts sensor coverage from 27% to 90%","feed_subtitle":"In a distributed network, each mobile sensor moves only when the local coverage gain pays for the energy.","key_machinery":"The load-bearing object is the Voronoi-based decomposition of coverage into local contributions, specifically Lemma 2's identity (13) asserting that the mobile-sensor-covered regions are disjoint and jointly exhaust the area covered by mobile sensors, so that maximizing each local Fi also maximizes the global objective F. The gradient computation (15)-(16) converts the change in a local region into three computable integrals, and the step-size rule (18) plus the epsilon threshold convert that gradient into guaranteed, energy-conscious movement. The projection step keeps each sensor inside its assigned region and in line of sight.","core_discovery":"Using the Elfes probabilistic sensing model, the paper defines network coverage as an integral of the maximum detection probability, then uses Voronoi partitioning to rewrite it as the sum of local coverage terms. The central formula is the gradient of a mobile sensor's local coverage (Theorem 1, equations 15-16): a surface integral over the overlap of the sensor's sensing disk with its local region, a line integral over the moving arc of the sensing boundary, and additional line integrals for obstacle-vertex cut-offs. All of these rely only on neighbor positions and obstacle geometry. Algorithm 1 then performs distributed gradient ascent on this local term with a dynamic step size, projecti","pith_inferences":["The authors leave implicit that the step-size rule (18) depends on the gradient norm in the exponent, so sensors in tiny local regions near obstacle vertices may take very different step sizes than sensors in open areas; a direct numerical comparison with the alternative rule they mention (ηt = min(ηmax, η0/(φ(xi)||∇Fi||))) would test whether the chosen rule is the better trade-off.","Lemma 2 assumes the mobile local regions are disjoint even when two mobile sensors whose sensing ranges overlap are not communication neighbors. A simple two-sensor counterexample with communication radius smaller than the overlap distance would show whether the right-hand side of (13) equals the true coverage F; if not, the local gradient is climbing a surrogate objective.","The visibility constraint appears in D(si) but is not used in Lemma 1's proof, so a point q that is nearest to an occluded sensor but visible to a farther one is a place where the assigned sensor probability differs from the true maximum; the authors could quantify this discrepancy in a one-obstacle simulation."],"forward_implications":["If the decomposition and gradient ascent are sound, each mobile sensor can decide its motion using only information from within communication range, so the strategy scales to large networks without a central planner.","Because movement is accepted only when the local coverage gain exceeds epsilon, the algorithm naturally stops when marginal gains are small, saving energy and giving a concrete stopping rule.","The obstacle terms in the gradient and the projection step together let sensors route around obstacles while preserving existing communication links.","Starting from a connected configuration, the total coverage never decreases and the algorithm terminates in finite time, giving predictable deployment behavior.","In the paper's simulation with 30 mobile and 5 stationary sensors, the rule raises area coverage from about 27% to 90%, indicating the practical magnitude of improvement."],"supporting_citations":[{"why":"Supplies the heterogeneous-network gradient technique with obstacles and the boundary-term structure that this paper extends.","marker":"[10]"},{"why":"Provides the gradient-based coverage optimization method and the Leibniz-rule derivation used for Theorem 1.","marker":"[12]"},{"why":"Introduces the Elfes probabilistic sensing model used throughout to define sensing probability.","marker":"[13]"},{"why":"Establishes the Voronoi-distributed coverage control framework that motivates the local-region formulation.","marker":"[9]"},{"why":"Supplies the Leibniz integral rule used in the proof of the gradient formula.","marker":"[29]"},{"why":"Supports the convergence argument in Remark 3 via non-decreasing coverage and local-maximum stopping.","marker":"[8]"}],"fun_headline_variants":["Distributed gradient rule lifts sensor coverage to 90%","Sensor network self-optimizes coverage via distributed gradients","Local-gradient rule lets sensors self-optimize coverage","Coverage improves to 90% via distributed gradient deployment"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the coverage decomposition (13) is exact: the local mobile sensing regions are disjoint and jointly exhaust the area covered by mobile sensors, and the nearest-sensor identity in Lemma 1 holds even when obstacles block line of sight. If overlap or visibility breaks this decomposition, the gradient climbs a surrogate objective rather than the true coverage.","fun_headline_variants_meta":{"raw":{"variants":["Distributed gradient rule lifts sensor coverage to 90%","Sensor network self-optimizes coverage via distributed gradients","Local-gradient rule lets sensors self-optimize coverage","Coverage improves to 90% via distributed gradient deployment"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001436,"raw_usage":{"total_tokens":5581,"prompt_tokens":654,"completion_tokens":4927,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":398,"completion_tokens_details":{"reasoning_tokens":4871}},"tokens_in":398,"tokens_out":4927,"duration_ms":38344,"temperature":1.0,"reasoning_tokens":4871,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T11:21:34.346163+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small network with two mobile sensors whose sensing ranges overlap but whose communication radius is smaller than the distance between them, so they are not neighbors, plus one stationary sensor. Compute the true global coverage F directly by integrating φ(q)·max_i psi(q), then compute the right-hand side of equation (13); any positive discrepancy shows the decomposition and therefore the gradient's target undercounts coverage. Alternatively, place one obstacle so that the closest sensor to a point q is occluded while a farther sensor has line of sight: at that point ps(q) > psi(q) for","supporting_citations":[{"cited_title":"A distributed strategy to maximize coverage in a heterogeneous sensor network in the presence of obstacles,","cited_arxiv_id":null,"evidence_quote":"Supplies the heterogeneous-network gradient technique with obstacles and the boundary-term structure that this paper extends."},{"cited_title":"A gradient-based coverage optimization strategy for mobile sensor networks,","cited_arxiv_id":null,"evidence_quote":"Provides the gradient-based coverage optimization method and the Leibniz-rule derivation used for Theorem 1."},{"cited_title":"Derivatives of probability functions and some applica- tions,","cited_arxiv_id":null,"evidence_quote":"Supplies the Leibniz integral rule used in the proof of the gradient formula."},{"cited_title":"Distributed deployment algorithms for coverage improvement in a network of wireless mobile sensors: Relocation by virtual force,","cited_arxiv_id":null,"evidence_quote":"Supports the convergence argument in Remark 3 via non-decreasing coverage and local-maximum stopping."}],"review_version":1}