{"id":"c15a9694-5735-444d-bcff-046636d72332","arxiv_id":"1909.00329","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a single-antenna over-the-air computation system, the paper gives a closed-form optimal transmit-receive policy and proves the average computation error decays as O(1/√K).","lead":"Over-the-air computation lets many wireless sensors send simultaneously so a receiver directly computes their sum. This paper derives the optimal transmit and receive scaling under each sensor's peak power limit, and shows the average computation error decays like one over the square root of the number of sensors while average power use also tends to zero.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Appendix B's Lemma 3a is false, leaving the claimed scaling-law proofs in Theorems 3 and 4 and Proposition 5 unproved, though the O(1/sqrt(K)) upper bound for the optimal policy survives.","rationale":"I read the central claim as Theorem 1 (closed-form optimal switching policy) plus Proposition 4 (optimal-policy average MSE decays as O(1/sqrt(K))). I found no independent defect in Theorem 1: the interval decomposition, the lemma sequence, and the unimodality argument are internally consistent. Proposition 4's O(1/sqrt(K)) upper bound survives the Appendix B flaw, since it can be obtained from the feasible first-sqrt(K) policy and the valid inverse-gamma bound in (A.4). The false Lemma 3a is nevertheless a genuine gap in the paper's advertised scaling-law and energy-efficiency results, exactly the reader's weakest assumption, so the conditional verdict should stand. I would not move to accept, because the appendix flaw is real and material to several results, nor to reject, because the central optimization result and the O(1/sqrt(K)) optimal-policy upper bound are not contradicted by anything I found.","tokens_in":21274,"tokens_out":16061,"duration_ms":152093,"concrete_test":"Compute E[X_(K-1)/X_K] for K=100 using Renyi's representation: draw 10^6 realizations of K i.i.d. Exp(1) variables, sort, and average the ratio. The lemma's bound predicts a value at most about 0.02, while the actual value is near 0.8. If this check confirms the failure, recompute (A.11) and (A.15) with the correct ratio expectations: the adjacent-k terms are O(1) rather than O(1/K), so the claimed asymptotic forms in Theorem 3(3) and Theorem 4(3) require a different argument and should not be reported as proved until that argument is supplied.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Appendix B's Lemma 3a states E[X_(i)/X_(j)] <= (i+1)(K-i+1)/(K(j-2)) for all j>2. This is false. Set i=K-1 and j=K. Renyi's representation gives X_(K) = X_(K-1) + Z_K with Z_K ~ Exp(1), while X_(K-1) grows like log K in mean. The ratio X_(K-1)/X_K therefore tends to 1 in probability, and for K=100 its expectation is about 0.8, whereas the lemma's bound is about 2/K. The false bound is the only support for the lower-bound parts of Theorem 3 (cases 2 and 3, equations (A.9)-(A.13)) and for the upper-bound part of Theorem 4 (case 3, equations (A.15)-(A.16)); Proposition 5 and the claimed rigorous existence of a policy that is both computation-effective and energy-efficient rest on these. The central O(1/sqrt(K)) decay claim for the computation-optimal policy (Proposition 4) is not endangered, because it follows from the valid upper bound in (A.4) applied to the feasible first-sqrt(K) policy. The scaling-law statements should be re-proved with a corrected order-statistics bound or explicitly weakened.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a single-antenna over-the-air computation (AirComp) system with K sensors and one fusion center, where each sensor has a peak power constraint. The central optimization problem is to minimize the computation mean-squared error by jointly choosing the transmit and receive scaling factors. The authors derive a closed-form \"computation-optimal policy\": a switching policy with critical index i* = arg max_i g_i, where g_i is defined in Eq. (11); sensors with the smallest channel gains transmit at full power, while the remaining sensors use channel inversion. They also compare this MSE-of-sum AirComp problem with a sum-of-MSE remote-estimation MAC problem, characterizing the optimal sum-of-MSE policy and the achievable MSE region. The second half of the paper analyzes ergodic performance under Rayleigh fading, defining computation-effective and energy-efficient policies and deriving scaling laws for average MSE and average power consumption as K grows. The main scaling claims are an O(1/sqrt(K)) decay of average MSE for the computation-optimal policy, and the existence of a first-i policy with i(K) ~ sqrt(K) that is both computation-effective and energy-efficient.","tokens_in":21517,"tokens_out":7623,"duration_ms":127867,"significance":"If the main claims are established, the paper makes a useful contribution: the closed-form optimal Tx-Rx policy for the single-antenna AirComp problem is a clean result that improves on earlier suboptimal or numerical designs, and the comparison with the classical sum-of-MSE MAC problem is conceptually valuable. The O(1/sqrt(K)) upper bound on the optimal policy's average MSE is obtained by constructing an explicit feasible policy, so that part of the scaling analysis is credible and falsifiable. However, the more detailed scaling laws for first-i policies and the claimed existence of a simultaneously computation-effective and energy-efficient policy rest on an order-statistics lemma in Appendix B that is false as stated. These scaling-law claims are a large part of the paper's novelty, so the manuscript requires substantive correction rather than minor polishing.","major_comments":[{"comment":"Lemma 3a is false as stated. For i = K-1 and j = K, the claimed bound gives E[X_(i)/X_(j)] <= 2/(K-2), which is O(1/K). Rényi's representation gives X_(K) = X_(K-1) + Z_K with Z_K ~ Exp(1), while X_(K-1) = Theta(log K); therefore X_(K-1)/X_K tends to 1 in probability, and since this ratio lies in [0,1], its expectation also tends to 1, not 0. This lemma is used in equations (A.9)-(A.12) to derive the lower bounds in Theorem 3 and in equations (A.15)-(A.16) to derive the upper bound in Theorem 4. Consequently Theorem 3 cases 2 and 3, Proposition 3, and Proposition 5 are not established as written. The O(1/sqrt(K)) upper bound for the computation-optimal policy in Proposition 4 survives, because it follows from the valid bound in (A.4) applied to a feasible first-sqrt(K) policy. The scaling-law statements should be re-proved with a correct order-statistics argument, or explicitly weakened.","section":"Appendix B, Lemma 3a and Theorem 3, Appendix C and Theorem 4"},{"comment":"The proofs of Lemmas 2a, 2b, and 2c are omitted with the statement that they 'can be verified using the similar steps.' These lemmas are load-bearing: Theorem 1's optimality proof relies on the unimodality of the sequence {g_i} and the consequent unimodality of {MSE_i}. Please provide complete proofs of these lemmas, or cite a source that contains them, so that the main closed-form optimality claim is fully supported.","section":"Section III-A, Lemmas 2a-2c"},{"comment":"Proposition 5 claims that the first-sqrt(K) policy is energy-efficient with a decay rate between O(1/sqrt(K)) and O(log(K)/sqrt(K)). This specific quantitative claim depends on Theorem 4(3), whose upper bound uses Lemma 3a in (A.15). Since Lemma 3a is false, the stated decay rate for the power consumption is unproved. At present the paper only rigorously proves that this policy is computation-effective; the energy-efficiency claim should either be re-proved with a valid argument or presented as a numerical observation.","section":"Section V-C, Proposition 5"}],"minor_comments":[{"comment":"The figure caption says 'The average critical number of the computation-optimal policy versus the number of sensors,' but the text describes Fig. 8 as showing the average power consumption. The caption and the text should be made consistent.","section":"Section VI, Fig. 8"},{"comment":"The caption reads 'the average power consumption versus K,' while the surrounding text discusses the average computation MSE achieved by the multi-antenna policies. One of the two is mislabeled.","section":"Section VI, Fig. 9"},{"comment":"The line 'K-k+1 > K-k-17' appears to contain a typographical error; the subsequent replacement of (k+1) by k and the change from K-k+1 to K-k need a clearer derivation, not an unexplained numeric constant.","section":"Appendix B, Eq. (A.10)"},{"comment":"There are several small language and typographical errors, e.g., 'standard derivations' where 'standard deviations' is meant, and 'the the average' in the discussion of Fig. 9. A careful proofreading pass is needed.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The false order-statistics lemma is a genuine technical error that invalidates several of the scaling-law theorems, but the main optimization result and the constructive O(1/sqrt(K)) upper bound appear sound and repairable. I recommend major revision rather than rejection, provided the authors correct the lemma or honestly weaken the affected claims. The paper is transparent about the parallel work in [22], and I do not see a circularity or attribution problem."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's what I'd want you to know before reading: the optimization half is solid, the scaling-law half is not. The paper gives a closed-form MSE-minimizing Tx-Rx policy for single-antenna AirComp under per-sensor peak power: a switching policy where the i* sensors with the weakest channels transmit at full power and the rest do channel inversion. I read through the interval-partition argument and the unimodality lemmas; it holds together. This policy is not new - Cao et al. [22] got it independently - but the derivation here is self-contained, and the authors say so, which is honest.\n\nWhat is new is the scaling analysis for Rayleigh fading: the first-i policies, the claimed 1/sqrt(K) decay, the computation-effectiveness vs energy-efficiency tradeoff, and the sum-of-MSE vs MSE-of-sum comparison. The comparison result (Theorem 2, Remark 3) is correct and useful. The O(1/sqrt(K)) upper bound on the optimal policy's average MSE is also correct, because it comes from an explicit feasible policy (first-sqrt(K)) and the upper bound in (A.4), which does not depend on the bad lemma.\n\nThe soft spot is Appendix B. Lemma 3a claims E[X_(i)/X_(j)] <= (i+1)(K-i+1)/(K(j-2)) for exponential order statistics. That is false. For i=K-1, j=K the right side is about 2/K, while Renyi's representation gives X_(K) = X_(K-1) + Exp(1), so X_(K-1)/X_(K) -> 1 in probability and its expectation is bounded below by a constant. The lemma feeds directly into the lower-bound parts of Theorem 3 (cases 2 and 3) and the upper-bound part of Theorem 4 (case 3). With the lemma gone, those scaling-law statements and Proposition 5 are not proved. They may be true - some of them look plausible - but the paper as written doesn't establish them. Also, the conclusion says the optimal policy has vanishing average power consumption, but that's only numerical evidence, no proof.\n\nWho is this for: people working on AirComp or analog function computation who want the closed-form optimal policy and a design rule for large K. They should read the first four sections and then treat the appendix with suspicion until the authors supply a corrected order-statistics bound or weaken the claims. This deserves a serious referee; I'd send it out, but with a request to fix or restrict the appendix before publication. My own verdict would be conditional: accept after the scaling-law proofs are repaired.","headline":"Clean closed-form AirComp policy and a survivable O(1/sqrt(K)) upper bound, but the appendix's Lemma 3a is false and takes down most of the scaling-law proofs.","tokens_in":22076,"tokens_out":4380,"would_cite":true,"duration_ms":39397,"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":"Closed-form switching policy minimizes over-the-air computation error","keywords":["over-the-air computation","computation mean-squared error","transmit-receive scaling policy","peak power constraint","Rayleigh fading","scaling laws","order statistics","wireless sensor networks"],"falsifier":"Compute $\\mathbb{E}[X_{(K-1)}/X_{(K)}]$ for $K=100$ i.i.d. standard exponentials using Rényi's representation: the bound in Lemma 3a gives about $2/K = 0.02$, while numerical evaluation gives a constant near 0.8. If this discrepancy is confirmed, the proof of Theorems 3 and 4 does not support the stated scaling laws.","tokens_in":1764,"feed_emoji":"📡","tokens_out":5802,"duration_ms":143923,"temperature":0.7,"pith_summary":"The paper solves in closed form the problem of minimizing the computation mean-squared error of an over-the-air computation (AirComp) system with K single-antenna sensors and one receiver, under a peak power budget per sensor. It proves that the optimal transmit-receive policy is a switching policy: sort the channels, transmit at full power on the weakest channels up to a cutoff index, and invert the remaining channels, with the cutoff chosen by maximizing a scalar sequence $g_i$. Under Rayleigh fading, it shows the average per-sensor computation error of this optimal policy decays as $O(1/\\sqrt{K})$ as K grows, so adding many sensors keeps the per-sensor error small; numerical results also show the average per-sensor power consumption vanishes. This gives a concrete design rule for data aggregation in IoT and a benchmark for AirComp analysis.","feed_headline":"Closed-form switching policy minimizes over-the-air computation error","feed_subtitle":"Weakest sensors transmit at full power, the rest invert the channel, and the error decays as 1/sqrt(K).","key_machinery":"The carrying object is the sequence $g_i$ and the partition of the positive real line into intervals $\\mathcal{S}_i = (1/(h_{i+1}\\sqrt{P}), 1/(h_i\\sqrt{P})]$. On each interval, the optimal transmit gains have a switching form: the $i$ weakest sensors use full power and the rest use channel inversion, which turns the joint transmit-receive MSE into a quadratic function of the receiver gain $a$. The sequence $g_i$ is unimodal, so $i^*$ equal to the maximizer of $g_i$ selects the globally optimal interval. This reduces a non-convex joint optimization to a one-dimensional comparison of K known quantities.","core_discovery":"Ordering channel coefficients as $h_1 \\le \\cdots \\le h_K$, define $g_i = \\sqrt{P} \\sum_{k=1}^i h_k / (\\sigma^2 + P \\sum_{k=1}^i h_k^2)$. The global optimum of the non-convex MSE-minimization problem is described by $i^* = \\arg\\max_{1 \\le i \\le K} g_i$: sensors with the $i^*$ smallest channels transmit at peak power $\\sqrt{P}$, the remaining sensors transmit with channel-inversion gains $1/(a^* h_k)$, and the receiver scaling $a^*$ is chosen by minimizing the resulting quadratic MSE in that interval. Under Rayleigh fading, this policy is computation-effective, with average per-sensor MSE at most $O(1/\\sqrt{K})$, and simulations indicate it is also energy-efficient in that its average per-sensor power tends to zero. The related sum-of-MSE estimation policy for a traditional MAC is full-power transmission by every sensor, and it equals the AirComp optimal policy precisely when $g_K = \\max_i g_i$.","pith_inferences":["If the order-statistics inequality in Lemma 3a is false, the exact decay-rate claims for first-$\\imath$ policies would need repair, but the optimal policy's $O(1/\\sqrt{K})$ upper bound could survive because it is obtained via a feasible first-$\\sqrt{K}$ policy.","The switching structure depends only on the channel ordering and the quadratic MSE, so a similar closed-form rule should generalize to other fading distributions and correlated channels, with only the scaling constants changing.","Deriving the distribution of the optimal cutoff $i^*$ would turn the numerically observed energy efficiency of the optimal policy into an analytic theorem."],"forward_implications":["The optimal transmit-receive policy is given by a one-line rule depending only on the sorted channel gains; no iterative or alternating optimization is needed.","With Rayleigh fading, the optimal policy achieves average per-sensor MSE at most $O(1/\\sqrt{K})$, so larger sensor populations improve per-sensor computation accuracy.","Numerical results indicate the optimal policy is simultaneously computation-effective and energy-efficient, a combination neither benchmark achieves.","The optimal MAC sum-of-MSE policy is full-power transmission by all sensors, and it coincides with the AirComp policy exactly when $g_K = \\max_i g_i$.","The channel-inversion benchmark has infinite average per-sensor MSE under Rayleigh fading, while full-power transmission has non-vanishing average per-sensor MSE."],"supporting_citations":[{"why":"First posed the AirComp transceiver optimization problem and provides the suboptimal multi-antenna baseline that the single-antenna closed-form solution improves upon.","marker":"[12]"},{"why":"Parallel work that independently derived the same single-antenna computation-optimal policy with a different proof structure.","marker":"[22]"},{"why":"Supplies the order-statistics representations and distributions of exponential random variables used in the scaling-law proofs.","marker":"[26]"},{"why":"Provides the Rayleigh fading channel model used for the ergodic performance analysis.","marker":"[25]"},{"why":"Develops robust analog function computation over multiple-access channels, establishing the AirComp foundations.","marker":"[4]"},{"why":"Introduces the idea of harnessing interference for analog function computation in wireless sensor networks.","marker":"[5]"},{"why":"Supplies the channel-inversion benchmark policy and the wireless-powered AirComp setting.","marker":"[13]"},{"why":"Provides the CEO-problem remote-estimation context that motivates the traditional MAC sum-of-MSE comparison.","marker":"[15]"}],"fun_headline_variants":["AirComp MSE decays as 1/sqrt(K) with closed-form optimal policy","Closed-form AirComp policy: weak sensors max power, others invert","Scaling law for AirComp: error O(1/sqrt(K)), energy efficient","Optimal AirComp transmission: closed-form solution to non-convex MSE","AirComp: weakest sensors transmit full power, error decays"],"cache_read_input_tokens":24192,"weakest_assumption_plain":"The scaling-law proofs rely on an order-statistics inequality (Lemma 3a) bounding ratios of ordered exponential channel gains; if that inequality fails, the exact decay-rate claims are not established.","fun_headline_variants_meta":{"raw":{"variants":["AirComp MSE decays as 1/sqrt(K) with closed-form optimal policy","Closed-form AirComp policy: weak sensors max power, others invert","Scaling law for AirComp: error O(1/sqrt(K)), energy efficient","Optimal AirComp transmission: closed-form solution to non-convex MSE","AirComp: weakest sensors transmit full power, error decays"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000354,"raw_usage":{"total_tokens":1988,"prompt_tokens":1071,"completion_tokens":917,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":687,"completion_tokens_details":{"reasoning_tokens":820}},"tokens_in":687,"tokens_out":917,"duration_ms":12648,"temperature":1.0,"reasoning_tokens":820,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:57:21.346255+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $\\mathbb{E}[X_{(K-1)}/X_{(K)}]$ for $K=100$ i.i.d. standard exponentials using Rényi's representation: the bound in Lemma 3a gives about $2/K = 0.02$, while numerical evaluation gives a constant near 0.8. If this discrepancy is confirmed, the proof of Theorems 3 and 4 does not support the stated scaling laws.","supporting_citations":[{"cited_title":"MIMO over-the-air computation for high-mobility multimodal sensing,","cited_arxiv_id":null,"evidence_quote":"First posed the AirComp transceiver optimization problem and provides the suboptimal multi-antenna baseline that the single-antenna closed-form solution improves upon."},{"cited_title":"Optimized Power Control for Over-the-Air Computation in Fading Channels","cited_arxiv_id":"1906.06858","evidence_quote":"Parallel work that independently derived the same single-antenna computation-optimal policy with a different proof structure."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the order-statistics representations and distributions of exponential random variables used in the scaling-law proofs."},{"cited_title":"Inference from randomized transmissions by many backscatter sensors,","cited_arxiv_id":null,"evidence_quote":"Provides the Rayleigh fading channel model used for the ergodic performance analysis."},{"cited_title":"Robust analog function computation via wireless multiple-access channels,","cited_arxiv_id":null,"evidence_quote":"Develops robust analog function computation over multiple-access channels, establishing the AirComp foundations."},{"cited_title":"Harnessing interference for analog function computation in wireless sensor networks,","cited_arxiv_id":null,"evidence_quote":"Introduces the idea of harnessing interference for analog function computation in wireless sensor networks."},{"cited_title":"The CEO problem,","cited_arxiv_id":null,"evidence_quote":"Provides the CEO-problem remote-estimation context that motivates the traditional MAC sum-of-MSE comparison."}],"review_version":1}