{"id":"884ca147-9530-4a6a-9418-65086940da1b","arxiv_id":"1908.08827","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The asymptotic degree distribution of the typical vertex in inhomogeneous and passive random intersection graphs holds under just a finite first moment.","lead":"This note proves that the typical vertex degree distribution in two random intersection graph models converges to a known limit under only a first moment condition, resolving two open conjectures. The proof is a short truncation argument that separates everyday nodes from rare heavy ones.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof's final 'Letting M→∞' step asserts the truncated limit laws converge to d* without proof; this is the least-secure link, though routine to verify.","rationale":"The central theorems are almost certainly correct. The truncation inequalities (3)-(5) are valid, and the tail contribution E check X1 is o(1) under Theorem 1's EX1<∞ and Theorem 2's (i)-(ii) via the standard uniform-integrability/truncation argument. The note's proof, however, has a second implicit limit: the family of truncated limit laws hat d*(M) must converge to d*. This is stated but not demonstrated. It is routine—characteristic functions converge because the Poisson means and the relevant size-biased/mixed-Poisson distributions converge—so I do not regard it as a correctness risk. It is, though, the least-supported step in the logic, and a referee could reasonably ask for two lines of justification. The 'minimal/optimal' terminology is not a mathematical objection. Overall the reader's conditional verdict stands; no change is needed.","tokens_in":3257,"tokens_out":24302,"duration_ms":262542,"concrete_test":"Verify the missing step by computing the probability generating function of hat d* in both models and showing it converges to that of d*. For the passive model, write φ_M(s)=exp(β^{-1}E[Z1{Z≤M}](E[s^{ξ_M}]-1)), where ξ_M is size-biased of Z1{Z≤M}; show E[s^{ξ_M}]→E[s^{ξ}] and the mean tends to EZ, so φ_M(s)→φ(s) for each s∈[0,1). For the inhomogeneous model, use dominated convergence to show P(Λ_i^{(M)}=r)→P(Λ_i=r) for each fixed r and EΛ_i^{(M)}→EΛ_i, then the claim distributions converge. Matching pgfs for all s settles the sandwich step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the sandwich argument, after (6) the paper needs P(hat d*≥k) → P(d*≥k) as M→∞. This is not automatic from the statements of Theorems A/B and is merely asserted ('Letting M→+∞ we obtain...'). For the passive model, hat d* is a compound Poisson limit whose Poisson mean is β^{-1} E[Z 1{Z≤M}] and whose claim distribution is the size-biased distribution of Z 1{Z≤M}; for the inhomogeneous model, hat d* is a compound Poisson limit with mixed-Poisson ingredients built from X1 1{X1≤M}. The convergence of these laws to the target d* does follow from the first-moment conditions: dominated convergence gives E[X1 1{X1>M}]→0, E[Z 1{Z≤M}]→EZ, and, for each fixed r, the size-biased probabilities and mixed-Poisson masses converge. But this verification is absent from the note. If for some k this convergence failed, inequality (6) would not pin liminf and limsup to P(d*≥k), and the distributional limit would not be established. Hence the central theorem is supported by a genuine but fillable gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies the asymptotic degree distribution of a typical vertex in two random intersection graph models: the inhomogeneous random intersection graph and the passive random intersection graph. Building on two earlier theorems that require a finite second moment for the inhomogeneous model and a 4/3 moment condition for the passive model, the author proves that the same compound-Poisson limits hold under only a finite first moment. The proof truncates the underlying weights at level M, applies the earlier theorems to the bounded truncated variables, bounds the contribution of large weights by its mean, and then passes M to infinity to recover the unrestricted limit.","tokens_in":3481,"tokens_out":9917,"duration_ms":93938,"significance":"If the two theorems are correct, they resolve the conjectures stated in [3] and [2] and are the expected optimal-moment results for these models. The truncation idea and the sandwich inequality (3) are simple and valid, and the paper is concise. The main caveat is that the final passage M→∞ in the limiting laws is asserted rather than proved; this is a routine but necessary verification. With that gap filled, the paper would be a genuinely useful short contribution to the literature on degree distributions in random intersection graphs.","major_comments":[{"comment":"The assertion \"Letting M→+∞ we obtain P(ˆd*≥k)→P(d*≥k)\" is not justified. Theorems A and B are applied at a fixed truncation level M, and their statements do not by themselves imply convergence of the limiting laws as M grows. In the inhomogeneous case one must show that the mixed-Poisson parameters, namely a_1(M)=E[X_1 1_{X_1≤M}] and Λ_2(M)=X_1 1_{X_1≤M} b_1 β^{-1/2}, together with the resulting size-biased claim probabilities, converge to their untruncated counterparts; in the passive case one must show that the Poisson mean β^{-1}E[Z 1_{Z≤M}] and the size-biased law of Z1_{Z≤M} converge to β^{-1}EZ and the size-biased law of Z. These facts follow by dominated convergence from the first-moment conditions, but the verification is absent. Since Eq. (6) only bounds liminf and limsup by P(ˆd*≥k), the final distributional limit is not established until this convergence is supplied.","section":"Section 3, after Eq. (6)"},{"comment":"The sentence \"conditions of Theorem 1 (Theorem 2) imply E ˇX_1 = o(1)\" is too compressed for the passive model. This is a uniform-integrability statement, and it should be proved rather than merely asserted. It follows from conditions (i) and (ii) of Theorem B, i.e. convergence in distribution together with convergence of first moments, but the paper does not say so. The same kind of verification is needed when applying Theorem B to the truncated variables for each fixed M: one should check the convergence of E[(X_1 1_{X_1≤M})^{4/3}] and choose M avoiding possible atoms of Z. Because the bound on P(ˇd≥1) in (5) is load-bearing for the limsup in (6), this missing justification should be supplied.","section":"Section 3, Eqs. (4)-(5)"}],"minor_comments":[{"comment":"There are two typos: \"inhomogenious\" should be \"inhomogeneous\", and \"a simply and elegant proof\" should be \"a simple and elegant proof\".","section":"Introduction"},{"comment":"The formula for P(τ_1=r) should be written as (r+1)/EΛ_2 to avoid the ambiguous reading \"r + 1/EΛ_2\".","section":"Eq. (2)"},{"comment":"References [6] and [10] are the same paper by Jaworski and Stark (2008) and should be merged or cross-referenced; reference [13] contains the typo \"Electronical\" for \"Electronic\".","section":"References"},{"comment":"The notation ˆX_i and ˇX_i is clear from context, but a one-line comment explaining the natural coupling (so that ˆd+ˇd equals the original degree) would make inequality (3) easier to verify for a reader.","section":"Section 3"}],"recommendation":"major_revision","confidential_remarks":"This is a very short note whose central idea is correct in spirit. The missing dominated-convergence verification is routine, and I would not recommend rejection. I would ask the authors to add a short lemma or paragraph proving convergence of the truncated compound-Poisson laws, and to justify the o(1) tail statement for the passive model, before the paper is accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take on Bloznelis' note. It does what it says: removes the second-moment condition in Theorem A and condition (iii) in Theorem B, settling conjectures from [3] and [2]. The proof is a truncation sandwich: replace each X_i by X_i 1{X_i≤M} and X_i 1{X_i>M}, show the degree of the truncated graph is close to the original, apply the earlier theorems to the truncated graph, and let M→∞. That is the right strategy, and the inequalities in (3)–(5) check out. The tail bounds are simple union bounds, and the moment assumptions do make the tail probabilities tend to zero. For the inhomogeneous model this is just EX_1 <∞; for the passive model the convergence-in-distribution plus convergence of means gives uniform integrability, so E X_1 1{X_1>M}→0. The core argument is sound.\n\nThe real weakness is the sentence 'Letting M→∞ we obtain ...' in Section 3. The paper does not justify that the truncated limit law \\hat d^*(M) converges to d^* as M→∞. This is not immediate from Theorems A and B; it needs a short argument with dominated convergence on the Poisson mean and the size-biased claim probabilities. It is a routine check and the first-moment condition gives it, but for a note whose entire contribution is this limiting step, it should be written out. Right now the proof has a genuine but fillable gap.\n\nTwo smaller things. The abstract and introduction say 'minimal'/'optimal' moment conditions. The paper proves sufficiency, not necessity; there is no lower-bound argument showing the first moment is required. That is worth fixing, because the word 'optimal' is doing more work than the proof does. And reference [6] is duplicated as [10] — same paper by Jaworski and Stark. Editorial, but embarrassing.\n\nWho is this for? Specialists in random intersection graphs. It is not a big new framework, and outside that area the impact is modest. But the conjectures were explicit, the proof is short and mostly rigorous, and the gap is small. I would send it to a referee who works in the area; the referee can check the dominated-convergence step in ten minutes. With that paragraph added and the 'optimal' wording toned down, I'd accept it.","headline":"A short, correct truncation proof resolves two conjectures on random intersection graph degree distributions; the final limit step is compressed but fillable.","tokens_in":3973,"tokens_out":1878,"would_cite":true,"duration_ms":17665,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C80","05C07","05C82"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the asymptotic degree distribution of a typical vertex in inhomogeneous and passive random intersection graphs converges to the same compound-Poisson limit under only a finite first-moment condition, confirming two…","keywords":["degree distribution","random intersection graph","inhomogeneous random intersection graph","passive random intersection graph","compound Poisson limit","power law","heavy tails","minimal moment conditions"],"falsifier":"Simulate the passive model with $m/n \\to \\beta$, attribute sizes with a heavy tail such as $\\Pr(X_1 = k) \\sim c k^{-5/2}$ (finite mean, infinite $4/3$ moment), and compare the empirical degree distribution with the compound Poisson law whose Poisson mean is $\\beta^{-1}\\mathbb{E} Z$ and whose summands follow the size-biased distribution of $Z$. A persistent mismatch as $n,m$ grow would refute Theorem 2; the specific quantity to monitor is whether $\\mathbb{E}[X_1 \\mathbf{1}\\{X_1 > M\\}]$ tends to zero uniformly.","tokens_in":3051,"feed_emoji":"🕸️","tokens_out":16405,"duration_ms":145394,"temperature":0.7,"pith_summary":"Random intersection graphs model networks in which two vertices are linked whenever they share at least one attribute from a common pool, and they are studied because they reproduce heavy-tailed degree distributions and clustering seen in real networks. This note determines the asymptotic degree distribution of a typical vertex in two such models, the inhomogeneous and the passive random intersection graph, and shows that a finite first moment of the attribute-size variable is the only moment condition needed. The paper proves two conjectures: the earlier second-moment condition for the inhomogeneous model and the 4/3-moment condition for the passive model are both redundant. If the claims are right, the limiting degree law—a compound Poisson sum in which the number of contributing attributes is Poisson and each contribution is drawn from a size-biased distribution—holds in the full heavy-tailed regime where only the mean attribute size is finite.","feed_headline":"Finite first moment settles degree limits in two graph models","feed_subtitle":"Two conjectures confirmed: heavy-tailed degree distributions need only a finite first moment.","key_machinery":"The argument is carried by the truncation sandwich inequality $\\hat d \\le d \\le \\hat d + \\check d$, where $d$ is the degree of the distinguished vertex, $\\hat d$ its degree in the graph built only from the capped variables $\\hat X_i = X_i \\mathbf{1}\\{X_i \\le M\\}$, and $\\check d$ its degree in the graph built from the tail variables $\\check X_i = X_i \\mathbf{1}\\{X_i > M\\}$. This reduces the distributional limit to two facts: on the truncated graph the old theorems apply because the variables are bounded, and the tail graph is asymptotically edge-free, with $P(\\check d \\ge 1) \\le b_1 \\sqrt{m/n}\\, \\mathbb{E}\\check X_1$ for the inhomogeneous model and $P(\\check d \\ge 1) \\le (n/m)\\, \\mathbb{E}\\check X_1$ for the passive model. Since the moment assumptions imply $\\mathbb{E}\\check X_1 = o(1)$ as $M \\to \\infty$, the sandwich forces the degree distribution to coincide with the limiting compound Poisson law from the truncated model.","core_discovery":"The paper's central claim is Theorem 1: the existing limit theorem for inhomogeneous random intersection graphs remains true when the condition $\\mathbb{E} X_1^2 < \\infty$ is weakened to $\\mathbb{E} X_1 < \\infty$, so the degree of a typical vertex still converges in distribution to $d^* = \\sum_{j=1}^{\\Lambda_1} \\tau_j$ with the compound-Poisson structure defined by the size-biased Poisson law. Its second central claim is Theorem 2: the passive-model limit theorem remains true when condition (iii), which requires $\\mathbb{E} Z^{4/3} < \\infty$ and convergence of the $4/3$ moments, is dropped; convergence of $X_1$ in distribution to $Z$ together with convergence of the first moment suffices, and the limit is again the compound Poisson variable $\\sum_{j=1}^{\\Lambda} \\widetilde{Z}_j$ with size-biased summands. Both results are proved by the same truncation sandwich: split each attribute size at a level $M$, apply the old theorems to the bounded part, and show that the large part contributes no edges in the limit because $\\mathbb{E}[X_1 \\mathbf{1}\\{X_1 > M\\}] \\to 0$.","pith_inferences":["Because the proof only uses first-moment control of the tail, the same truncation sandwich should extend to other local statistics of these graphs, such as the number of triangles through a vertex or the local clustering coefficient, where earlier arguments assumed higher moments; this is a testable extension, not a claim of the paper.","For the passive model the proof effectively replaces the $4/3$-moment condition with uniform integrability of $X_1$, which follows from convergence in distribution plus first-moment convergence; one could therefore restate the theorem in terms of uniform integrability and drop the auxiliary variable $Z$.","A quantitative version of the result is within reach: under regular variation of the tail of $X_1$, the rate at which the degree distribution approaches its limit should be controlled by $\\mathbb{E}[X_1 \\mathbf{1}\\{X_1 > M\\}]$, giving explicit convergence rates that the paper does not derive.","The sandwich argument is generic enough to transfer to any random intersection graph variant whose edge probability is monotone in the product $X_i Y_j$, provided the large-part graph is small in expectation; neighbouring models with tunable clustering are natural candidates."],"forward_implications":["The inhomogeneous random intersection graph has the same limiting degree law under $\\mathbb{E} X_1 < \\infty$ alone, so attribute distributions with infinite second moments no longer fall outside the theory.","The passive model's degree limit holds under convergence in distribution plus first-moment convergence, with no condition on higher moments and no separate moment-convergence assumption.","Both conjectures from earlier work are settled: the extra moment conditions in the two old theorems are now known to be unnecessary.","The tail part of the attribute distribution is asymptotically irrelevant to the typical degree, so the limiting law is determined entirely by the truncated, small-to-moderate part of the attribute sizes.","When the limiting attribute variable has a power-law tail, the size-biased summand distribution is also power-law, so the degree limit exhibits the heavy-tailed behavior associated with these models."],"supporting_citations":[{"why":"It supplies the passive-model limit theorem (Theorem B) and the conjecture that its condition (iii) is redundant, which Theorem 2 proves.","marker":"[2]"},{"why":"It supplies the inhomogeneous-model limit theorem (Theorem A) and the conjecture that its second-moment condition can be weakened, which Theorem 1 proves.","marker":"[3]"},{"why":"It defines the passive random intersection graph model whose typical degree is studied in Theorem 2.","marker":"[5]"},{"why":"It introduces random intersection graphs, the general model family whose degree asymptotics are analyzed.","marker":"[11]"},{"why":"It defines the inhomogeneous random intersection graph model used in Theorem 1.","marker":"[13]"}],"fun_headline_variants":["First moment suffices for degree limits in intersection graphs","Two graph models: degree limit holds with just finite mean","Heavy tails need only finite first moment for degree limits","Minimal condition: finite first moment gives degree distribution"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof collapses if the average size of the discarded large part of the attribute variable fails to shrink to zero as the truncation level grows, because then the large-part subgraph could keep creating edges no matter how high the cutoff is set.","fun_headline_variants_meta":{"raw":{"variants":["First moment suffices for degree limits in intersection graphs","Two graph models: degree limit holds with just finite mean","Heavy tails need only finite first moment for degree limits","Minimal condition: finite first moment gives degree distribution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000498,"raw_usage":{"total_tokens":2368,"prompt_tokens":799,"completion_tokens":1569,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":415,"completion_tokens_details":{"reasoning_tokens":1504}},"tokens_in":415,"tokens_out":1569,"duration_ms":11207,"temperature":1.0,"reasoning_tokens":1504,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:29:10.551667+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the passive model with $m/n \\to \\beta$, attribute sizes with a heavy tail such as $\\Pr(X_1 = k) \\sim c k^{-5/2}$ (finite mean, infinite $4/3$ moment), and compare the empirical degree distribution with the compound Poisson law whose Poisson mean is $\\beta^{-1}\\mathbb{E} Z$ and whose summands follow the size-biased distribution of $Z$. A persistent mismatch as $n,m$ grow would refute Theorem 2; the specific quantity to monitor is whether $\\mathbb{E}[X_1 \\mathbf{1}\\{X_1 > M\\}]$ tends to zero uniformly.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the passive-model limit theorem (Theorem B) and the conjecture that its condition (iii) is redundant, which Theorem 2 proves."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the inhomogeneous-model limit theorem (Theorem A) and the conjecture that its second-moment condition can be weakened, which Theorem 1 proves."},{"cited_title":"Godehardt and J","cited_arxiv_id":null,"evidence_quote":"It defines the passive random intersection graph model whose typical degree is studied in Theorem 2."},{"cited_title":"Karo´nski, E","cited_arxiv_id":null,"evidence_quote":"It introduces random intersection graphs, the general model family whose degree asymptotics are analyzed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It defines the inhomogeneous random intersection graph model used in Theorem 1."}],"review_version":1}