{"id":"b4ee10cc-e89b-4aef-9889-b78f14696faf","arxiv_id":"1908.07375","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The SINR percolation model with random powers is shown to have both a subcritical and a supercritical phase under weaker conditions on the power distribution than previously known.","lead":"This thesis proves phase transitions for a wireless network model where each user transmits with a random, potentially unbounded power. It extends prior percolation results that assumed bounded powers to unbounded, integrable powers with exponential moments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Step 3 of Theorem 4.5 relies on a false implication: a gδ-path between high-power points does not put them within distance δ, and the proof gives no SINR path through low-power relays.","rationale":"Both the strict-inequality boundary and the Step 3 false implication concern the same renormalization bridge, but the false implication is the more load-bearing one: even after replacing '≥' by '>' in condition (a), the proof as written does not construct SINR edges between adjacent high-power sites. The reader's weakest_assumption highlighted the boundary issue; the rationale also noted the false implication, so my agreement is partial. The underlying result is plausibly correct and likely repairable by a standard blocking argument with a smaller lattice spacing, so the conditional verdict stands rather than moving to acceptance or rejection. The proposed check would determine whether the repair is routine or whether the theorem needs additional hypotheses.","tokens_in":46888,"tokens_out":12661,"duration_ms":143334,"concrete_test":"Test Step 3 directly: take a configuration of two adjacent sites that satisfy Definition 4.6, place the high-power points at far corners of Q1(z), Q1(z'), and connect them by a δ-chain of low-power points inside Q6(z). Compute the SINR bound for the claimed γ* on the direct pair; it will fail when their distance exceeds δ. Then check whether a repair exists by scaling the lattice spacing to s<δ/(2√d) and reproving Proposition 4.8 for cubes of side s using the same hypotheses (a)-(b). If the scaled proof goes through, the theorem survives in modified form; if not, the central claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing gap is in Section 4.3, Step 3. After defining good sites (Definition 4.6), the proof chooses Xi∈Q1(z), Xj∈Q1(z') with ρi,ρj>r and uses A(z)=A(z')=1 to assert that Xi and Xj are connected by a path in gδ(Xλ)∩Q6(z) and therefore |Xi−Xj|≤δ. The second implication is false: a path of many δ-edges does not make its endpoints within δ. Moreover, even if one replaces the implication by the actual path, the intermediate vertices are not guaranteed to have power >r, so the SINR lower bound (which assumes transmitter power >r) cannot be applied to each edge. Thus the chain from lattice percolation to SINR percolation is missing. A separate boundary issue noted by the reader compounds this: at r=N0τ/l(0), δ=0, so Definition 4.6 cannot hold for large λ, and Proposition 4.8's limit is false; this is fixable by requiring strict inequality, but the Step 3 gap is not fixed by that change. As written, Theorem 4.5 is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This thesis studies continuum percolation in three multihop wireless-network models. Chapters 2 and 3 are expository, covering the Boolean model following Meester–Roy and the SINR model for Cox point processes following Tóbiás. The original contribution is Chapter 4, which introduces the SINR model with random powers on a Poisson point process and claims a phase transition: a subcritical phase when the path-loss function decays as l(r) ≤ r^{-β} and the power tail satisfies P(ρ>r) ≤ r^{-α} with αβ > 2d−1 (Corollary 4.4), and a supercritical phase when P(ρ>r)>0 for some r ≥ N0τ/l(0) and E[e^{αρ}]<∞ (Theorem 4.5). The subcritical proof reduces to the Boolean model with random radii, while the supercritical proof follows the three-step renormalization scheme of Dousse et al. The paper is written as a Master's thesis, with large parts devoted to background material.","tokens_in":47062,"tokens_out":11878,"duration_ms":125462,"significance":"If Theorem 4.5 and Corollary 4.4 are established, the paper would generalize the bounded-power results of Kong and Yeh to unbounded random powers having a positive probability of exceeding a threshold, in dimensions d≥2; the subcritical condition is natural and not previously stated in this generality. The expository parts are standard but well organized, and the subcritical part of Chapter 4 appears sound. However, as detailed below, the supercritical proof of Theorem 4.5 contains a load-bearing gap, so the main positive claim is not yet established as written. The paper's machine-checkable proofs are not present; the arguments are conventional pen-and-paper proofs with citations to external theorems (Meester–Roy, Tóbiás, Dousse et al.), and the subcritical reduction uses those theorems correctly.","major_comments":[{"comment":"The proof asserts that if two high-power points Xi∈Q1(z) and Xj∈Q1(z') are connected by a path in gδ(Xλ)∩Q6(z), then |Xi−Xj|≤δ. This implication is false: a path consisting of edges of length at most δ can have endpoints arbitrarily far apart within Q6(z). Moreover, the intermediate vertices along such a path are not known to have power >r, so the SINR lower bound (which requires transmitter power greater than r) cannot be applied to every edge of the path. Consequently, percolation of the lattice does not imply an infinite path in the SINR graph as currently argued, and Theorem 4.5 is not established. The transfer argument needs to be reworked, for example by constructing connectivity through the marked high-power process rather than through the full unmarked Poisson process.","section":"Section 4.3, Step 3 (after Definition 4.6)"},{"comment":"The theorem assumes r ≥ N0τ/l(0), but the proof requires the strict inequality r > N0τ/l(0). Indeed, δ is defined by δ = l^{-1}(N0τ/r)/2; at r = N0τ/l(0), strict decrease of l gives δ=0, the graph gδ has no edges between distinct Poisson points almost surely, and condition (2) of Definition 4.6 cannot hold for large λ. In that case Proposition 4.8 fails. The theorem statement should assume r > N0τ/l(0), and the presentation should note that the equality case is excluded.","section":"Section 4.3, Theorem 4.5(a) and Definition 4.6"},{"comment":"The proof asserts lim_{λ→∞} P(A(o)=0)=0 without a quantitative argument, and the definition of 'good' connects all points of Xλ inside Q3(z), including low-power points. Even if the limit statement were proved, it would not provide the high-power connectivity needed in Step 3, because the SINR transfer requires a path whose vertices all have power above r. This is closely related to the first comment, but it indicates that the renormalization step must be reformulated at the level of the marked high-power process; the current Definition 4.6 is not sufficient for the claimed conclusion.","section":"Section 4.3, Proposition 4.8 and Definition 4.6"}],"minor_comments":[{"comment":"In the SINR inequality, the symbol ϵ appears where r is meant: 'ρi l(...) > ϵ l(δ)' should read 'ρi l(...) > r l(δ)'.","section":"Section 4.3, Step 3"},{"comment":"The phrase 'at least 7 d distant' is ambiguous; the independence argument should specify a numerical separation such as 13^d or 7^d and justify the lower bound m ≥ N/7^d on the size of the independent subset.","section":"Section 4.3, Proposition 4.8"},{"comment":"The conditions on the path-loss function in Chapter 4 (continuous, strictly decreasing on supp l, 1≥l(0), integrable) differ from those in Chapter 3, where additionally l(0)>τN0/P is assumed; this distinction should be stated explicitly so that the reader understands why rB is not used in Chapter 4.","section":"Chapter 4 model definition"},{"comment":"The abstract says that a supercritical phase exists if the random power has a 'large enough essential supremum', but condition (a) of Theorem 4.5 is a probability condition P(ρ>r)>0; the wording should be aligned with the theorem.","section":"Abstract and Introduction"},{"comment":"There are numerous typos and grammatical errors (for example, 'finitly', 'proced', and the equation numbering in Chapter 3 is occasionally inconsistent); the manuscript needs a careful language and formatting pass before publication.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"This is a Master's thesis; Chapters 2 and 3 are largely expository, and the original contribution is Chapter 4. The subcritical result appears correct, but the supercritical theorem has a genuine gap in the Step 3 transfer, which is load-bearing. I recommend major revision; the authors should either repair the supercritical proof (for instance, by working with a high-power marked process) or clearly restrict the claims to what is actually proved. If the manuscript is intended as a journal paper, it would also benefit from substantial condensation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: Chapter 4 is the reason to look at this. Loeffler weakens Kong-Yeh's bounded-power assumption for SINR percolation on PPPs to nonnegative integrable powers, treats d≥2, and gives a subcritical phase under tail decay. The subcritical proof is a sound reduction to Meester-Roy: connections in the SNR graph are contained in a Boolean model with random radii l^{-1}(N0τ/ρ), and the moment computation with αβ>2d−1 checks out. The expository chapters on the Boolean model and on Tóbiás's Cox SINR work are also solid. The supercritical proof, Theorem 4.5, is where the trouble is. The lattice construction has a boundary problem: Definition 4.6 sets δ = l^{-1}(N0τ/r)/2, and if r equals N0τ/l(0), then δ=0 and the good-site condition collapses. That alone is fixable by requiring r>N0τ/l(0). But Step 3 has a deeper gap. Good sites guarantee a path in the Gilbert graph gδ between points Xi∈Q1(z) and Xj∈Q1(z'); they do not imply |Xi−Xj|≤δ. A path of δ-edges can have long total length. And even if one replaces that false implication by the actual path, the intermediate vertices are arbitrary Poisson points; their powers need not exceed r, so the SINR lower bound, which needs transmitter power >r, does not apply. The proof gives no route through low-power relays. This is the load-bearing transfer from lattice percolation to SINR percolation, and as written Theorem 4.5 is not established. I do not read this as dishonesty. The author states conditions carefully, attributes building blocks correctly (Meester–Roy, Tóbiás), and the subcritical half is genuinely correct. But the supercritical claim needs serious repair: define good sites via a percolating network of high-power points, or prove a path whose every transmitter has power >r, and handle the boundary case r=N0τ/l(0). The typos—ε rather than r in the final displayed calculation, and the wrong inequality direction in Proposition 4.8—are minor by comparison. Who should read it: people working in continuum percolation for interference-limited networks, and anyone supervising a thesis student through a first original proof. It deserves a serious referee, not a desk reject; with the Step 3 gap fixed it would be a serviceable extension of Kong–Yeh. I would accept it for review but ask for a rewritten supercritical section.","headline":"The new supercritical theorem in Chapter 4 is plausible but the written proof has a real gap in Step 3; the subcritical half is clean and the thesis is worth engaging, but a referee should insist on fixing the transfer.","tokens_in":761,"tokens_out":821,"would_cite":false,"duration_ms":41318,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","60D05","60G55"],"pacs":[],"model":"deepseek-v4-flash","headline":"A SINR network whose users transmit with random powers percolates at large user intensities whenever the power can exceed a large threshold with positive probability and has finite exponential moments.","keywords":["continuum percolation","SINR model","random transmission powers","Poisson point process","phase transition","interference","shot noise","Gilbert graph"],"falsifier":"Test the boundary value $r = N_0\\tau/\\ell(0)$. With $\\delta = 0$, the Gilbert graph $g_0$ has no edges between distinct points, so any unit cube containing two Poisson points is not good; for large $\\lambda$ such configurations occur with probability close to 1, making $\\mathbb{P}(A(o)=0) \\to 1$, the opposite of Proposition 4.8's conclusion. A simulation or exact computation of $\\mathbb{P}(A(o)=0)$ as $\\lambda \\to \\infty$ at this boundary would settle whether the theorem's non-strict statement can be saved.","tokens_in":46611,"feed_emoji":"📡","tokens_out":12912,"duration_ms":120188,"temperature":0.7,"pith_summary":"The paper studies a telecommunication network in which users are placed by a Poisson point process in $\\mathbb{R}^d$, $d \\ge 2$, and each user transmits with an independent random power. It asks when an infinite multi-hop communication component exists in the graph where two users are linked if their signal-to-interference-and-noise ratio exceeds a threshold in both directions. It proves a subcritical phase when the path-loss function decays like $r^{-\\beta}$ and the power tail decays like $r^{-\\alpha}$ with $\\alpha\\beta > 2d-1$: for small intensity there is no infinite component. It proves a supercritical phase when the power exceeds $N_0\\tau/\\ell(0)$ with positive probability and some exponential moment is finite: for all sufficiently large intensities there is a positive interference tolerance $\\gamma^*(\\lambda)$ such that percolation holds for every $\\gamma \\le \\gamma^*(\\lambda)$. The result removes the bounded-power restriction of earlier work and extends the model from $\\mathbb{R}^2$ to all dimensions $d \\ge 2$.","feed_headline":"Random signal powers still percolate past a density threshold","feed_subtitle":"Removes bounded-power restriction and proves the phase transition for interference-limited networks in every dimension d≥2.","key_machinery":"The proof constructs a site percolation on $\\mathbb{Z}^d$. A site $z$ is good if some Poisson point in the unit cube near $z$ has random power above $r$, and all Poisson points in a larger cube are connected inside an even larger cube by the Gilbert graph with radius $\\delta = \\ell^{-1}(N_0\\tau/r)/2$; a site is open if it is good and its weighted shot noise $\\tilde{I}_6(z) = \\sum_i \\rho_i \\ell_6(|z-X_i|)$ is at most $M$. The bridge from lattice to continuum is the SINR lower bound: two points at distance at most $\\delta$ with powers above $r$ and total interference at most $M$ satisfy $\\rho\\ell(\\delta)/(N_0+\\gamma M) > \\tau$ whenever $\\gamma \\le \\gamma^* = (N_0/M)(\\ell(\\delta)/\\ell(2\\delta)-1)$, since $\\ell(2\\delta) = N_0\\tau/r$. Thus percolation of open lattice sites forces an infinite connected component in the SINR graph.","core_discovery":"The central claim is Theorem 4.5. In the SINR model with random powers on a Poisson point process of intensity $\\lambda$ in $\\mathbb{R}^d$, $d \\ge 2$, assume the nonnegative power variable $\\rho$ is not a.s. zero, has finite expectation, satisfies $\\mathbb{P}(\\rho > r) > 0$ for some $r \\ge N_0\\tau/\\ell(0)$, and has finite exponential moment $\\mathbb{E}[e^{\\alpha\\rho}] < \\infty$ for some $\\alpha > 0$. Then for every sufficiently large $\\lambda$ there exists $\\gamma^*(\\lambda) > 0$ such that the SINR graph percolates for every $\\gamma \\le \\gamma^*(\\lambda)$. Together with Corollary 4.4, which gives a subcritical phase when $\\ell(r) \\le r^{-\\beta}$ and $\\mathbb{P}(\\rho > r) \\le r^{-\\alpha}$ with $\\alpha\\beta > 2d-1$, this yields a phase transition under conditions that allow unbounded, heavy-tailed powers.","pith_inferences":["A likely repair for the boundary case is to replace $r \\ge N_0\\tau/\\ell(0)$ by $r > N_0\\tau/\\ell(0)$: whenever $\\mathbb{P}(\\rho > r) > 0$ holds at the boundary, continuity from below yields a slightly larger threshold with positive mass, so the auxiliary radius $\\delta$ can be made positive.","The exponential-moment condition in Theorem 4.5 enters only through the control of the marked Poisson shot noise; a natural testable extension is whether a finite $(2d-1+\\epsilon)$-th moment of $\\rho$ would suffice if the interference estimate is handled by truncation instead of the moment-generating function.","The subcritical criterion $\\alpha\\beta > 2d-1$ suggests a phase diagram in the two exponents: for fixed dimension, percolation at low density is governed only by the product of the power-tail exponent and the path-loss exponent, not by either alone. Plotting or simulating the critical intensity against $\\alpha$ and $\\beta$ would test whether the product form is sharp."],"forward_implications":["An unbounded random power distribution can be used in the supercritical regime, provided its tail has enough positive mass at large powers and its exponential moment is finite; this removes the earlier boundedness condition on $\\rho$.","For large user densities, the network tolerates a strictly positive interference level $\\gamma^*(\\lambda)$ while still containing an infinite communication component, so the phase transition in $\\gamma$ occurs at a positive value.","The quantitative subcritical criterion $\\alpha\\beta > 2d-1$ links the decay of the path-loss function and the decay of the power tail; both can be heavy as long as their product decays fast enough.","The statements hold for every dimension $d \\ge 2$, so the $d=2$ restriction of the earlier bounded-power model is removed."],"supporting_citations":[{"why":"Studied the same SINR model with random powers for Poisson point processes in $\\mathbb{R}^2$ under bounded-power assumptions; the paper's Theorem 4.5 weakens these conditions and extends to $d \\ge 2$.","marker":"[KY07]"},{"why":"Introduced the signal-to-interference ratio percolation graph and the three-step lattice-mapping strategy (good/closed sites, Peierls argument, transfer to continuum) that the supercritical proof follows.","marker":"[DFM+06]"},{"why":"Provides the Cox-point-process SINR framework and the finite constant $K_0$ bounding sums of the shifted path-loss function, used in Proposition 4.9's interference control.","marker":"[T´ob19]"},{"why":"Supplies the Boolean model with random radii phase-transition theorem that underlies the subcritical phase proof in Proposition 4.1.","marker":"[MR96]"}],"fun_headline_variants":["Unbounded signal powers still percolate in ad-hoc networks","Heavy-tailed powers don't block wireless percolation","SINR percolation survives unbounded random powers","Phase transition for SINR with integrable random powers","Percolation persists when signal powers are heavy-tailed"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The supercritical-phase proof needs a power threshold strictly above $N_0\\tau/\\ell(0)$ so that the auxiliary radius $\\delta = \\ell^{-1}(N_0\\tau/r)/2$ is positive; the theorem states only $r \\ge N_0\\tau/\\ell(0)$, and the boundary case is where the good-site construction loses its footing.","fun_headline_variants_meta":{"raw":{"variants":["Unbounded signal powers still percolate in ad-hoc networks","Heavy-tailed powers don't block wireless percolation","SINR percolation survives unbounded random powers","Phase transition for SINR with integrable random powers","Percolation persists when signal powers are heavy-tailed"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000347,"raw_usage":{"total_tokens":1988,"prompt_tokens":1121,"completion_tokens":867,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":737,"completion_tokens_details":{"reasoning_tokens":786}},"tokens_in":737,"tokens_out":867,"duration_ms":9308,"temperature":1.0,"reasoning_tokens":786,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:47:06.514091+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test the boundary value $r = N_0\\tau/\\ell(0)$. With $\\delta = 0$, the Gilbert graph $g_0$ has no edges between distinct points, so any unit cube containing two Poisson points is not good; for large $\\lambda$ such configurations occur with probability close to 1, making $\\mathbb{P}(A(o)=0) \\to 1$, the opposite of Proposition 4.8's conclusion. A simulation or exact computation of $\\mathbb{P}(A(o)=0)$ as $\\lambda \\to \\infty$ at this boundary would settle whether the theorem's non-strict statement can be saved.","supporting_citations":[],"review_version":1}