{"id":"95104d26-98e5-419b-a0fd-f8f9d1889cf9","arxiv_id":"2505.21456","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper derives mean and variance formulas plus Cantelli tail bounds for the conditional false-alarm probability in a Poisson interferer field, and a beta approximation for its distribution.","lead":"A radar surrounded by randomly placed interferers experiences a different false-alarm rate in every neighborhood, and this paper derives formulas and bounds for that random performance spread. It argues that radar thresholds should be set from percentile guarantees rather than average false-alarm rates, which matters for automotive and high-reliability radar design.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1's moment formulas are not well-defined: the PGFL weight g(x,y) is sign-changing and its exponent integral diverges at y=x, so Eqs. (6), (7), and the derived Cantelli/beta results lack a valid foundation.","rationale":"The paper aims to characterize the distribution of the conditional false-alarm probability in a Poisson field, a worthwhile problem, and the beta approximation idea is reasonable. However the central derivation, Theorem 1, rests on a PGFL reduction that is not valid under the stated assumptions. The singularity of g at y=x is not a removable technicality: in the paper's own d=1, α=2 configuration the integral inside the exponent diverges, so the closed forms in Eqs. (6) and (7) are not defined. I also noticed a sign mismatch between Eq. (4) and Eq. (3), which reinforces that Eq. (5) does not describe the quantity being analyzed. Because the Cantelli bound and beta approximation are built directly on these moments, the advertised percentile guarantees have no proven foundation. The numerical Monte Carlo curves may be reproducible for a finite window and could even match a correctly regularized version, but they do not establish the unregularized R^d formulas. This is a strong, specific correctness risk; rejecting the paper as stated is appropriate, with a clear path to repair by specifying a finite domain, minimum separation, or convergent regularization. I agree with the reader's identification of the undefined PGFL integral as the key weakness.","tokens_in":9613,"tokens_out":17382,"duration_ms":165585,"concrete_test":"Evaluate B_ε(x)=∫_{ε<|y−x|<1} |1 − y^2/(x^2−y^2)| dy for d=1, α=2. Since B_ε(x) ~ x ln(1/ε) as ε→0, the PGFL exponent in Eq. (6) is not Lebesgue-convergent, settling that Theorem 1 is undefined without an added regularization or a proved principal-value interpretation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III's central result, Theorem 1, obtains the mean and variance of the CFA probability by converting the sum-product functional in Eq. (5) into exponential PGFL expressions. The step marked (c) in Eq. (8) requires applying the probability generating functional of the PPP to the factor g(x,y)=y^α/(x^α−y^α). This is not a valid PGFL weight: it is sign-changing (positive for |y|<|x|, negative for |y|>|x|) and it has a non-integrable singularity on the sphere y=x. In the paper's own numerical setting d=1, α=2, |1−g(x,y)| ~ x/(2|y−x|) near y=x, so the integral of |1−g| over any neighborhood of the singularity diverges logarithmically; the same codimension-1 singularity occurs for all d (on |y|=|x| for general α). The standard PGFL for a PPP is formulated for nonnegative functions with finite ∫|1−v|, and no truncation, exclusion radius, or principal-value interpretation is stated anywhere in the paper. Additionally, g in Eq. (4) is the negative of the factor r_j^α/(r_j^α−r_i^α) appearing in Eq. (3), so Eq. (5) is not algebraically identical to the CFA expression being studied. Therefore Eqs. (6), (7), and (12) are formal at best, and the Cantelli bound (13) and beta approximation in Section III.A inherit this defect. Monte Carlo agreement in a finite simulation window (Fig. 2) cannot validate unregularized integrals over R^d. The approach may be repairable with a finite observation window or a minimum-separation regularization, but as written the paper's central claim is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an analytical framework for the distribution of the conditional false-alarm probability (CFA) and conditional detection probability (CD) in a radar system when interferers/clutters form a Poisson point process. The main contribution is Theorem 1, which claims closed-form expressions for the mean and variance of the CFA probability by applying higher-order Campbell-Mecke formulas and the probability generating functional (PGFL) of the PPP. On this basis the paper derives a Cantelli tail bound (Theorem 2), a beta-distribution approximation for the CFA, and analogous (mostly Markov) bounds for the CD probability. Numerical results compare the beta approximation with Monte Carlo simulations and discuss percentile-level radar design insights.","tokens_in":9982,"tokens_out":7729,"duration_ms":80365,"significance":"The problem addressed is relevant: moving from mean SINR or mean false-alarm metrics to distributional or meta-distribution guarantees is a genuine need for high-reliability radar and sensing systems. If the claimed moment formulas were correct, the paper would provide a useful design tool. The paper also clearly motivates the gap between average and percentile performance. However, the central contribution rests entirely on Theorem 1, and the derivation of Theorem 1 is invalid: the PGFL is applied to an inadmissible weight, and the integrals defining the moments are not well defined. The paper does not provide machine-checked proofs or code, and the Monte Carlo comparison cannot validate unregularized integrals over R^d. Hence the significance of the claimed results is not established.","major_comments":[{"comment":"The PGFL of a PPP is applied to the weight v(y)=y^alpha/(y^alpha-x^alpha). This weight is not a permissible PGFL argument: it is negative for |y|<|x|, larger than 1 for |y|>|x| when alpha is even, and it has a non-integrable singularity at y=x. Specifically, for d=1 and alpha=2, |1-v(y)| ~ x/(2|y-x|) near y=x, so the integral of |1-v| over any neighborhood of the singularity diverges logarithmically; for general d the same holds on the codimension-1 set |y|=|x|. The paper states no truncation, exclusion radius, or principal-value interpretation. Therefore the exponent -lambda integral x^alpha/(y^alpha-x^alpha) dy in Eq. (6) is not defined in the Lebesgue sense, and the subsequent formulas (7), (12), and the Cantelli bound (13) inherit this defect. Additionally, the sign is inconsistent with the standard PGFL identity E[prod v(y)] = exp(-lambda integral (1-v(y)) dy), which would give exp(+lambda integral x^alpha/(y^alpha-x^alpha) dy), not the exponent written in Eq. (8).","section":"III, Eqs. (3)-(5)"},{"comment":"The sum-product functional F in Eq. (5) is not algebraically equal to the CFA probability in Eq. (3). With g(x,y) = y^alpha/(x^alpha-y^alpha) as defined in Eq. (4), the product over j in Phi\\{x} of g(x,y_j) equals (-1)^{|Phi|-1} times the factor product r_j^alpha/(r_j^alpha-r_i^alpha) appearing in Eq. (3). Since the total number of interferers |Phi| is random, the parity factor (-1)^{|Phi|-1} does not cancel in expectation. Consequently, E[F] is not equal to p_FA, and Theorem 1's identification of Eq. (6) as the unconditional false-alarm probability is invalid. The paper also states that g maps to [0,infty), but g takes negative values, which compounds the issue.","section":"III, Theorem 1 and Eq. (12)"},{"comment":"The second-moment expression in Eq. (12) is formal for the same reasons as the first moment. The terms g^2(x,y) and g(x,y)g(x',y) are not nonnegative and have singularities at y=x and y=x' respectively, so the integrals appearing in the exponents (1 - g^2) and (1 - g g') are not defined without regularization. Moreover, the differential notation in the exponent of the second term is malformed: “exp (-lambda integral 1 - g(x,y)g(x',y) dy dx dx')” lacks the closing parenthesis and the measure structure for the outer double integral. No dominated convergence or integrability argument is given to justify exchanging the expectation over Phi with the sum/integral representations in Eq. (10)-(11). Thus the variance formula is not established.","section":"IV, Corollary 1 and Lemma 2"},{"comment":"The detection-probability bounds in Section IV rely on the same undefined integrals, for example the inner integral x^alpha/(y^alpha-x^alpha) dy in Corollary 1 and in Lemma 2. The paper does not provide any additional convergence conditions for these quantities. Since the CFA moment formulas are invalid, the derived bounds for the CD probability inherit the same unsupported foundation.","section":"V, Fig. 2"},{"comment":"The numerical validation in Fig. 2 reports agreement between the beta approximation and Monte Carlo simulations, but the simulation setup is not described: the intensity lambda, the spatial dimension d, the observation window, and the number of realizations are not stated. Simulations over a finite window cannot validate integrals taken over all of R^d, especially when those integrals diverge. The figure therefore does not provide evidence for the correctness of Eqs. (6), (7), or the beta approximation.","section":"V, Fig. 2"}],"minor_comments":[{"comment":"The codomain of g is stated as [0, infty), but g(x,y)=y^alpha/(x^alpha-y^alpha) takes negative values when y^alpha < x^alpha; the codomain should be R or the notation should be changed.","section":"Section II, Eq. (4)"},{"comment":"The notation r_i and r_j is used without definition; the reader must infer that r_i is the distance of point i from the origin. Please define these distances explicitly.","section":"Eq. (3)"},{"comment":"The expression for T2 has a missing parenthesis and an unclear differential structure: it should read exp(-lambda integral (1 - g(x,y)g(x',y)) dy) dx dx'.","section":"Eq. (12)"},{"comment":"There are several typos, including “detction” in the index terms and “reveals a new approach” repeated in the abstract; also “For an appropriate choice of d” in Section II.A is missing a period.","section":"Abstract and Index Terms"}],"recommendation":"reject","confidential_remarks":"The paper's central theorem is invalid: the PGFL step is an improper application to a singular, sign-changing weight, and the moment definitions are not well defined. The algebraic mismatch between Eq. (3) and Eq. (5) is a further independent error. The paper could potentially be revised with a finite observation window or a minimum-separation model, but that would constitute a substantial change of the model and derivations, not a minor revision. The numerical figures do not report the simulation parameters and cannot serve as evidence for the unregularized integrals. I recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's the short version: the paper's main result, Theorem 1, does not hold as written. The PGFL step in (8) is applied to a function g that is sign-changing and has a non-integrable singularity at y=x. Worse, the g defined in (4) is the negative of the factor that appears in the sum-product expression (3). So Eq. (5) is not algebraically identical to the CFA probability, and the moment formulas (6)-(7) are formal at best. The Cantelli bound and the beta approximation inherit the problem.\n\nThat said, the paper deserves credit for framing the right question. Moving from average ROC metrics to distribution-level guarantees for radar in a Poisson interference field is a real need, and this is the first attempt I know to bring sum-product functionals and higher-order Campbell-Mecke tools to the CFA meta-distribution. The Monte Carlo results in Fig. 2 show the beta fit tracks the empirical distribution well, and the system-design insight—design to percentiles rather than means—is sensible and clearly illustrated.\n\nThe flaws are serious, not cosmetic. The sign error alone invalidates the derivation. Even if g were defined with the right sign, the PGFL requires finite ∫|1-g|, and with g = y^α/(y^α−x^α) you get a logarithmic divergence at y=x in the paper's own d=1, α=2 setting. No truncation, exclusion radius, or principal-value interpretation is stated anywhere. The simulation agreement is for a finite window and cannot certify the unregularized integrals over R^d.\n\nThe good news is the approach is repairable. Correct g, restrict the process to a finite observation window or add a minimum separation, and the moment derivations can be rerun with controlled convergence. The beta approximation and the percentile-design point would likely survive.\n\nWho gets value? Stochastic geometry folks interested in radar meta-distributions, and radar engineers who care about reliability guarantees. I'd not cite it in its current form, but I would not desk-reject it—it deserves a referee who can push the author to fix the PGFL step. As for acceptance, it's a reject-as-stated, with a clear path to major revision.\n\nFor your reading group: maybe, if you want a concrete case of a PGFL being misapplied.\n\nBest.","headline":"The central theorem is built on a misdefined g and a divergent PGFL integral—the moment formulas and derived bounds are unsupported as stated, though the problem and the beta-fit idea are worth taking seriously.","tokens_in":10506,"tokens_out":7530,"would_cite":false,"duration_ms":72708,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60D05","60G55"],"pacs":[],"model":"deepseek-v4-flash","headline":"Closed-form false-alarm moments enable percentile radar ROC design","keywords":["stochastic geometry","conditional false-alarm probability","meta-distribution","sum-product functional","Campbell-Mecke theorem","Cantelli inequality","beta approximation","radar ROC"],"falsifier":"Compute the right-hand sides of (6) and (12) for a one-dimensional Poisson process with $\\alpha=2$ and a finite field truncation, shrinking the excluded neighbourhood of $y=x$; if the result does not converge as the neighbourhood shrinks, the moment formulas lack a well-defined value and the Cantelli bound of Theorem 2 is not formally operational.","tokens_in":9336,"feed_emoji":"📡","tokens_out":7236,"duration_ms":70306,"temperature":0.7,"pith_summary":"The paper claims that the conditional false-alarm probability of a radar operating in a Poisson field of interferers and clutters, though a random variable over network realizations, has its mean and variance given by closed-form integral expressions derived through higher-order Campbell-Mecke calculus. If the formulas hold, the entire tail distribution can be bounded with Cantelli's inequality and approximated with a beta distribution, so radar thresholds can be chosen for percentile-level guarantees rather than average behavior. The paper further claims that the conditional detection probability admits only a looser Markov bound in general, but a Cantelli bound when the signal power is deterministic. A sympathetic reader would care because mean ROC curves turn out to be unrepresentative: the paper's numerical example shows a large fraction of radar realizations lying far from the mean.","feed_headline":"Radar false-alarm distribution pinned by its first two moments","feed_subtitle":"Mean-only ROC hides which radars fail; closed-form moment formulas and Cantelli bounds expose the tail.","key_machinery":"The carrying object is the sum-product functional $F=\\sum_{x\\in\\Phi} f(x)\\prod_{y\\in\\Phi, y\\neq x} g(x,y)$, with $f(x)=\\exp(-(\\gamma-N_0)/(KP x^{-\\alpha}))$ and $g(x,y)=y^\\alpha/(x^\\alpha-y^\\alpha)$. The argument moves the expectation inside the sums and products using the second-order Campbell-Mecke theorem, removes the conditioning points by reduced Palm expectation under Slivnyak's theorem, and evaluates the products by the Poisson probability generating functional. This reduces the moments to deterministic integrals over the kernel functions, which are then combined through Cantelli's inequality and the $\\beta$ approximation.","core_discovery":"On the paper's own terms, the discovery is that the random conditional false-alarm probability can be written as a sum-product functional of the point process, and that the reduced Palm expectation together with the Poisson probability generating functional reduces the first and second moments of this functional to deterministic integrals over the kernel functions. The resulting mean coincides with the standard false-alarm probability, while the variance splits into a diagonal term and a pairwise term, both closed-form up to well-defined integrals. These moments immediately give a Cantelli-type upper bound on the probability that a realization's false-alarm rate exceeds a target, and a beta-distribution approximation matched to those moments tracks the simulated distribution in the central region and tails. For detection probability, the extra randomness from the fluctuating target cross section blocks the second moment in general, leaving a Markov bound; with deterministic signal power, the paper derives a Cantelli bound from the same machinery.","pith_inferences":["The paper does not address the divergence of the kernel integral in $g(x,y)$ for $\\alpha=2$, $d=1$; a natural check is whether adding a small exclusion radius around $y=x$ makes the moment integrals finite and how the Cantelli bound depends on that radius.","The same moment machinery should carry over to path-loss kernels without the singularity, such as bounded or multi-slope models, where the beta approximation could be validated against Monte Carlo to separate the method from the kernel choice.","A second-moment derivation for the Swerling-I detection probability might be attempted by applying the stochastic Fubini theorem to the double copies of the interference field, which would extend percentile-level guarantees from false alarm to detection."],"forward_implications":["Radar thresholds can be chosen so that a specified percentile of realizations, not just the mean, keeps the false-alarm rate below a target.","The beta approximation gives a way to estimate tail probabilities of the conditional false-alarm probability without extensive Monte Carlo simulation.","For deterministic signal power, the second moment of the conditional detection probability becomes available, yielding a Cantelli bound on detection tails; otherwise only a looser Markov bound applies.","Mean-based ROC operating points are not representative of network behavior, since the distribution of the false-alarm and detection probabilities can be spread widely around their averages."],"supporting_citations":[{"why":"states the second-order Campbell-Mecke theorem used to evaluate the second moment of the CFA probability","marker":"[17]"},{"why":"gives the phase-type complementary CDF that underlies the sum-product representation of the CFA probability","marker":"[18]"},{"why":"supplies the stochastic Fubini theorem used to interchange expectation and signal integration for the detection probability","marker":"[19]"},{"why":"provides the one-dimensional Poisson highway radar model whose geometry the numerical setting adopts","marker":"[4]"},{"why":"introduces the fine-grained radar detection analysis whose SINR meta-distribution approach the paper extends to conditional false alarm","marker":"[14]"},{"why":"defines the meta-distribution concept that the paper's CFA and CD distributional analysis employs","marker":"[15]"}],"fun_headline_variants":["Radar false-alarm distribution tamed by moments and Cantelli","Mean ROC hides radar outliers; moments give tail bounds","Poisson field ROC: closed-form moments bound the tail","From mean to percentile: moment bounds for radar ROC","Conditional ROC spread: Cantelli bounds from Poisson moments"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The derivation assumes that the integrals $\\int_{\\mathbb{R}^d} x^\\alpha/(y^\\alpha-x^\\alpha)\\,dy$ and the corresponding product expectations are finite, but for the paper's own case $d=1$, $\\alpha=2$ the integral diverges near $y=x$ in the Lebesgue sense and no truncation or principal-value rule is stated.","fun_headline_variants_meta":{"raw":{"variants":["Radar false-alarm distribution tamed by moments and Cantelli","Mean ROC hides radar outliers; moments give tail bounds","Poisson field ROC: closed-form moments bound the tail","From mean to percentile: moment bounds for radar ROC","Conditional ROC spread: Cantelli bounds from Poisson moments"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000203,"raw_usage":{"total_tokens":1379,"prompt_tokens":931,"completion_tokens":448,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":547,"completion_tokens_details":{"reasoning_tokens":367}},"tokens_in":547,"tokens_out":448,"duration_ms":4894,"temperature":1.0,"reasoning_tokens":367,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:27:53.455460+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the right-hand sides of (6) and (12) for a one-dimensional Poisson process with $\\alpha=2$ and a finite field truncation, shrinking the excluded neighbourhood of $y=x$; if the result does not converge as the neighbourhood shrinks, the moment formulas lack a well-defined value and the Cantelli bound of Theorem 2 is not formally operational.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"states the second-order Campbell-Mecke theorem used to evaluate the second moment of the CFA probability"},{"cited_title":"Characterization of phase-type dis tributions,","cited_arxiv_id":null,"evidence_quote":"gives the phase-type complementary CDF that underlies the sum-product representation of the CFA probability"},{"cited_title":"The stochastic fubini theorem revisited,","cited_arxiv_id":null,"evidence_quote":"supplies the stochastic Fubini theorem used to interchange expectation and signal integration for the detection probability"},{"cited_title":"Stochastic geometry methods for modeling automotive radar interference,","cited_arxiv_id":null,"evidence_quote":"provides the one-dimensional Poisson highway radar model whose geometry the numerical setting adopts"},{"cited_title":"A Fine-Grained Analysis of Radar Detection in V ehicular Networks,","cited_arxiv_id":null,"evidence_quote":"introduces the fine-grained radar detection analysis whose SINR meta-distribution approach the paper extends to conditional false alarm"},{"cited_title":"Meta distributions—part 1: Definition and examples,","cited_arxiv_id":null,"evidence_quote":"defines the meta-distribution concept that the paper's CFA and CD distributional analysis employs"}],"review_version":1}