{"id":"01d3dbd2-269e-453e-9f3c-c104b7257753","arxiv_id":"1908.03757","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A broad class of thresholded hidden-variable network models always gives scale-free degree exponent 2, and a Fermi-Dirac soft threshold allows tuning the exponent to any value above 2.","lead":"This paper generalizes a class of network models in which each node carries a hidden fitness value, proving that many different fitness distributions and connection rules all produce scale-free networks with degree exponent 2. It also introduces a soft threshold connection rule that allows the exponent to be tuned continuously, which is useful for generating networks that match real-world systems.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; Eq. 34 is asymptotically valid because Δ−H(x)→∞ for all contributing nodes when Δ=(1/β)ln N.","rationale":"I re-derived the soft-threshold integral and its asymptotic expansion. The reader correctly identified Eq. 34 as the least rigorous-looking step, but the approximation is actually controlled: for β∈(0,1) and Δ=(1/β)ln N, all nodes contributing to the tail have Δ−H(x) large, so the leading term in the exact integral matches Eq. 34 with vanishing relative error. The hard-threshold gamma=2 results are clean and internally consistent. The only substantive residual issue is that the paper states the validity condition as 'H(x)≪Δ' instead of the more precise 'Δ−H(x)≫1', and it does not provide an explicit error bound. These are minor exposition gaps, not load-bearing flaws. Thus the reader's conditional verdict remains reasonable, but no change in verdict is required.","tokens_in":11722,"tokens_out":32144,"duration_ms":350479,"concrete_test":"Compute the exact degree distribution numerically for N=10^6, β=0.5 and β=0.8 using Eq. 33 with Δ=(1/β)ln N: evaluate F(t) by numerical quadrature, invert k(H), and generate the Poisson-mixture degree distribution. Fit the tail slope over k∈[10, N^β/10]. If the fitted exponent is within 0.05 of 1+1/β while respecting finite-size fluctuations, the concern about Eq. 34 is conclusively settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest assumption targets the approximation in Eq. 34, but this concern does not land in the thermodynamic limit. For β∈(0,1) and the proposed size-independent choice Δ=(1/β)ln N, the maximum hidden variable among N i.i.d. exponential-like nodes is O(ln N) with high probability, so H(x) remains below Δ and, crucially, Δ−H(x) grows as (1/β−1)ln N →∞ for every node. Writing t=H(x)−Δ, the exact integral in Eq. 33 is F(t)=∫_t^∞ e^{-z}/(1+e^{-βz})dz. For t→−∞, the dominant contribution comes from z near t, where 1/(1+e^{-βz})=e^{βz}(1+O(e^{βz})), giving F(t)=e^{-(1−β)t}/(1−β)+subdominant corrections. Thus k(x)=Ne^{t}F(t)=N/(1−β)e^{βt}(1+o(1)), which is exactly Eq. 34 as an asymptotic leading term, with relative error vanishing when Δ−H(x)→∞. The degree range relevant to the power-law tail has k up to N^β, corresponding to Δ−H(x)=(1−β)ln N/β→∞, so the approximation is uniformly valid in the tail. Even a full Poisson-mixture calculation preserves the same tail exponent. The paper's wording 'H(x)≪Δ' is imprecise; the needed condition is Δ−H(x)≫1, which holds. The absence of an explicit error bound is a presentation issue, not a correctness flaw.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies hidden-variable network models with thresholded linking functions. It defines three classes of models (exponential-like, power-like, and mixed) in which the hidden-variable density is expressed through a monotone transform H, G, or M, and the linking probability is zero below a threshold Delta and otherwise equal to an arbitrary function f-tilde of the additive, multiplicative, or mixed combination of the transformed variables. For each class the authors show analytically that the degree distribution satisfies p(k) proportional to k^{-2} in the thermodynamic limit, independent of the post-threshold form of f-tilde, and they give explicit conditions (L_Delta = 1/N, K_alpha = 1/N, J_alpha = 1/N) for the degree distribution to be size independent. The second half of the paper replaces the hard threshold with a 'reversed' Fermi-Dirac function with stiffness parameter beta, derives an approximate exponent gamma approximately 1 + 1/beta for beta in (0,1) and gamma approximately 2 for beta > 1, and supports the predictions with simulations for a Weibull example. The paper also discusses a transformation rule that maps models in the three classes into one another.","tokens_in":12098,"tokens_out":20549,"duration_ms":213073,"significance":"The hard-threshold part of the paper is a clean and useful unification: it reduces the previously observed k^{-2} universality of threshold hidden-variable models (Refs. [10,21,22]) to a single change-of-variables argument, and it works for any post-threshold kernel f-tilde rather than only for the Heaviside step function. The analytic steps in Eqs. (14)-(27) are correct under the stated assumptions, and the explicit size-independence conditions are practical and falsifiable. The soft-threshold mechanism is intuitive and the predicted tunable exponent is potentially useful for applications. The main caveat is that the soft-threshold exponent is an asymptotic approximation rather than a rigorously bounded result, and the supporting simulations are single-size and without error bars; however, the paper is explicit about the approximate character of this part, and the asymptotic validity of Eq. (34) is plausible in the thermodynamic limit. These caveats are local and do not affect the hard-threshold conclusions.","major_comments":[],"minor_comments":[{"comment":"The mixed-class size-independence condition appears inconsistent with the stated definitions. With s = 1 + M(x) and t = 1 + M(y), the density becomes dt/t^alpha and the kernel is f-tilde(st), so the integral should be integral_Delta^infinity f-tilde(z) z^{-alpha} dz, not integral_Delta^infinity f-tilde(z)/(1+z)^alpha dz. If the intended integration variable is z = st - 1, then f-tilde is evaluated at z + 1 and the lower limit should be Delta - 1. Please correct Eq. (28) and clarify the definition of z.","section":"Eq. (28), mixed class"},{"comment":"Eq. (34) is derived by replacing 1/(1+e^{-beta z}) with e^{beta z} for z = H(x) - Delta much less than 0. The working condition is therefore not merely H(x) << Delta but Delta - H(x) >> 1. With the proposed choice Delta = (1/beta) ln N this is satisfied for all contributing nodes in the thermodynamic limit, so the asymptotic exponent in Eq. (35) is sound; nevertheless the stated condition should be reworded to avoid confusing readers.","section":"Eq. (34), soft thresholding"},{"comment":"The replacement of the lower integration limit by 0 for beta > 1 is an N-to-infinity statement for fixed beta, not a uniform-in-beta statement. For finite N and beta just above 1, the omitted negative-z contribution is of order N^{-(beta-1)}/(beta-1), which can be substantial (for example, N = 20000 and beta = 1.1 give roughly a 30 percent correction to the integral). Please state explicitly that Eq. (38) is an asymptotic result in the thermodynamic limit with beta fixed, and ideally provide a finite-size scaling check.","section":"Eqs. (37)-(38), beta > 1 branch"},{"comment":"The simulation points in Fig. 3 have no error bars and use a single network size N = 20000. Since the analytic curve is approximate, showing the same gamma versus 1/beta curve for two or three sizes would make the claimed tunable-exponent behavior substantially more convincing.","section":"Fig. 3 and numerical support"},{"comment":"In Eq. (5), 'f-tilde(x,y)' should be a function of H(x) + H(y). Eq. (11) contains garbled notation around the integral and the Jacobian and should be rewritten. The transformation rules in Eq. (12) use a tilde on x without definition; please clarify. In Table 1, the Weibull distribution appears in both the exponential-like and power-like rows with different arguments; a sentence explaining the shared notation would help.","section":"Notation and typos"}],"recommendation":"minor_revision","confidential_remarks":"This is a solid but incremental contribution. The hard-threshold results are correct and clearly presented, and the soft-threshold construction is reasonable for a physics journal even without a fully rigorous error analysis. The main things to fix are the mixed-class condition in Eq. (28), the wording of the approximation condition in Eq. (34), and the unqualified beta > 1 statement in Eqs. (37)-(38). I saw no problematic citation patterns or scope issues; the paper fits physics.soc-ph."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: this is a solid, useful paper. The hard-threshold part is rigorous and extends the known inverse-square law to a much broader class of linking functions and fitness distributions. The soft-threshold part is genuinely new and gives a simple way to generate scale-free networks with tunable degree exponents, but its analytic support is asymptotic, not exact, and a couple of statements oversell it slightly.\n\nThe hard-threshold derivations are clean and correct. The change-of-variable steps in Eqs. (14)–(19) and (21)–(26), plus the mixed class, are all valid under the stated assumptions. The size-independence conditions (L_Δ(f̃)=1/N, etc.) are practically useful, and the mapping between the exponential-like, power-like, and mixed classes via transformations of H, G, M is elegant and explains the apparent universality of earlier results. This part is publishable as is.\n\nThe soft-threshold construction is the real contribution. Replacing the step with a Fermi-Dirac function and deriving γ ≈ 1+1/β for β∈(0,1) (Eq. 35) is new, and the simulations in Fig. 3 support it. The worry that Eq. (34) is uncontrolled does not, on closer look, land: with Δ=(1/β)ln N, the quantity Δ−H(x) diverges for all nodes that contribute to the tail, so the asymptotic expansion is uniformly valid. The paper's phrase 'H(x)≪Δ' is sloppy; the real condition is Δ−H(x)≫1. That is a presentation issue, not a mathematical flaw. An explicit error bound would still be nice, but I would not block acceptance on it.\n\nMinor soft spots: Fig. 3 lacks error bars, and the agreement is visual. The abstract's 'arbitrary scaling exponent' overstates the proven range—the formula gives a continuum of exponents, but with sub-power corrections and a breakdown as β→0, so 'tunable' rather than 'arbitrary' is the accurate word. The paper cites the relevant prior work (Caldarelli, Masuda, Fujihara) properly, and the self-citations are not a problem. The derivations are self-contained and not fitted to data.\n\nWho should read this: anyone working with hidden-variable or fitness-based network models, especially if they need a simple generator for scale-free networks with a specified γ. It deserves a serious referee; I would send it to review. After minor revisions—clarify the asymptotic condition, add error bars or state the visual nature of the fit, soften 'arbitrary'—it should be accepted.","headline":"A solid, useful paper: rigorous γ=2 universality for broad thresholded classes, and a new soft-threshold construction with tunable exponents—supported asymptotically, with minor oversights.","tokens_in":12581,"tokens_out":3201,"would_cite":true,"duration_ms":30513,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["89.75.Hc"],"model":"deepseek-v4-flash","headline":"For a broad family of thresholded hidden-variable networks, the degree distribution always decays as $k^{-2}$; softening the threshold tunes the exponent.","keywords":["hidden variable model","scale-free networks","degree distribution","inverse square law","threshold networks","soft thresholding","Fermi-Dirac linking function","tunable scaling exponent"],"falsifier":"Simulate the exponential-like model with $\\rho(x)=3x^2e^{-x^3}$, the soft link $f_{\\beta,\\Delta}=1/(1+e^{-\\beta(H(x)+H(y)-\\Delta)})$, and $N=10^5$ for $\\beta=0.3$ with $\\Delta=(1/\\beta)\\ln N$. Measure the degree exponent over the top three decades of $k$ and compare it with $1+1/\\beta\\approx 4.33$; if the effective exponent varies with the fitting window or the tail deviates from a pure power law, the approximation in Eq. (34) fails. For the hard-threshold claim, pick an $f$ with divergent $L_\\Delta(\\tilde f)$, for instance $\\tilde f(z)=e^{z/2}$ above the cutoff, and check that the degree distribution is no longer $k^{-2}$.","tokens_in":11533,"feed_emoji":"🕸️","tokens_out":9036,"duration_ms":77644,"temperature":0.7,"pith_summary":"The paper studies hidden-variable network models, in which each node carries an intrinsic fitness drawn from a distribution and two nodes connect with a probability set by a linking function. It shows that if the linking function has a hard lower cutoff—zero below a threshold and an arbitrary function of the combined fitness above it—then, for three large families of fitness distributions (exponential-like, power-like, and mixed), the degree distribution always obeys $p(k)\\propto k^{-2}$ in the thermodynamic limit, regardless of the detailed form of the linking function above threshold. The only requirements are that a transformed integral such as $L_\\Delta(\\tilde f)=\\int_\\Delta^\\infty \\tilde f(z)e^{-z}dz$ converges and that the threshold parameter is chosen so this integral equals $1/N$, the condition for a size-independent degree distribution. The paper then relaxes the hard cutoff to a smooth sigmoid linking probability, and argues, with simulations, that the scaling exponent can be tuned continuously, with $\\gamma\\approx 1+1/\\beta$ for $0<\\beta<1$ and $\\gamma\\to 2$ as $\\beta\\to\\infty$. If correct, this gives a broad explanation of why many threshold-type fitness models produce inverse-square degree distributions, and a practical way to generate sparse scale-free networks with realistic exponents.","feed_headline":"Thresholded variable networks force a universal k^-2 degree law","feed_subtitle":"A general proof covers exponential, power, and mixed couplings; softening the cutoff tunes the exponent.","key_machinery":"The central machinery is the change of variables $z=H(x)+H(y)$ (or $z=G(x)G(y)$, or $z=(1+M(x))(1+M(y))$) that makes the expected-degree integral factorise. In the exponential-like class this gives $k(x)=N L_\\Delta(\\tilde f)e^{H(x)}$ with $L_\\Delta(\\tilde f)=\\int_\\Delta^\\infty \\tilde f(z)e^{-z}dz$; substituting into the random-variable rule $p(k)=\\rho(x)/k'(x)$ then makes $p(k)\\propto k^{-2}$ explicit. The same identity is carried by the constants $K_\\alpha(\\tilde f)$ and $J_\\alpha(\\tilde f)$ in the other two classes. For the tunable extension, the key object is the reversed Fermi-Dirac function $f_{\\beta,\\Delta}=1/(1+e^{-\\beta(H(x)+H(y)-\\Delta)})$, whose integral is approximated under $H(x)\\ll\\Delta$ by $\\int e^{(\\beta-1)z}dz$, yielding $k(x)\\propto e^{\\beta H(x)}$ and hence $p_\\beta(k)\\sim k^{-(1+1/\\beta)}$.","core_discovery":"The central claim is that the inverse-square decay is a robust, universal outcome of thresholding in the hidden-variable model. For fitness densities of the form $\\rho(x)=H'(x)e^{-H(x)}$ with link probability zero when $H(x)+H(y)\\le\\Delta$ and equal to a general $\\tilde f(H(x)+H(y))$ above it, the expected degree factorises as $k(x)=N L_\\Delta(\\tilde f)e^{H(x)}$, where $L_\\Delta(\\tilde f)=\\int_\\Delta^\\infty \\tilde f(z)e^{-z}dz$. Substituting into the degree-transformation formula gives $p(k)\\propto k^{-2}$ together with the size-independence condition $L_\\Delta(\\tilde f)=1/N$. The analogous calculation for the power-like and mixed classes, with multiplicative or mixed combined variables and integrals $K_\\alpha(\\tilde f)$ and $J_\\alpha(\\tilde f)$, yields the same exponent with the same structure. The paper also claims that replacing the hard cutoff by the sigmoid $f_{\\beta,\\Delta}=1/(1+e^{-\\beta(H(x)+H(y)-\\Delta)})$ produces, for $0<\\beta<1$, an approximate power law with exponent $\\gamma\\approx 1+1/\\beta$, approaching $\\gamma=2$ for large $\\beta$.","pith_inferences":["An implication the authors leave implicit: the universality of exponent 2 means empirical networks with degree exponent near 2 do not require power-law fitness distributions; any sufficiently regular fitness distribution paired with a thresholded kernel will produce that exponent.","The soft-threshold exponent $\\gamma\\approx 1+1/\\beta$ is derived from a tail approximation, so I would expect the measured exponent to drift with the fitting window at small $\\beta$ and finite $N$; a systematic simulation over window sizes would reveal the size of the correction.","Because the change-of-variable argument works for any monotone $H$, the same thresholding construction should transfer to geographic and hyperbolic hidden-variable models, where the variables enter through distances or angular coordinates.","A testable practical recipe follows: for any proposed linking probability with a lower cutoff, compute $L_\\Delta(\\tilde f)$; if it converges, the $k^{-2}$ law is predicted and the cutoff parameter needed for size independence is fixed by $L=1/N$."],"forward_implications":["In the thermodynamic limit, every hard-thresholded model in the three generalised classes produces a degree distribution proportional to $k^{-2}$, independent of the functional form of the linking function above the threshold.","The role of the linking function's above-threshold shape reduces to setting the constants $L_\\Delta(\\tilde f)$, $K_\\alpha(\\tilde f)$, or $J_\\alpha(\\tilde f)$; the size-independence condition is that the relevant constant equals $1/N$.","Softening the cutoff with a reversed Fermi-Dirac function $f_{\\beta,\\Delta}$ yields, for $0<\\beta<1$, a sparse scale-free network with approximate exponent $\\gamma\\approx 1+1/\\beta$, and $\\gamma\\to 2$ as $\\beta\\to\\infty$.","For $\\beta>1$, choosing $\\Delta\\simeq \\ln N$ keeps the degree distribution size-independent and close to $k^{-2}$; for $\\beta<1$, the matching choice is $\\Delta\\simeq (1/\\beta)\\ln N$.","The transformation rules between the exponential-like, power-like, and mixed classes mean a construction or result in any one class carries over to the others."],"supporting_citations":[{"why":"Introduces the threshold hidden-variable model whose exponential-fitness plus step-function case is the baseline the paper generalises.","marker":"[10]"},{"why":"Supplies the continuous approximation $k(x)=N\\int f\\rho\\,dy$ and the degree-transformation formula $p(k)=\\rho(x)/k'(x)$ used throughout.","marker":"[6]"},{"why":"Establishes the analytic hidden-variable framework and the size-independence condition for correlated random networks.","marker":"[18]"},{"why":"Earlier demonstration that threshold graphs with intrinsic vertex weights show the $k^{-2}$ degree decay the paper extends.","marker":"[21]"},{"why":"Provides extreme-value-theory analysis of universal power laws in threshold networks, the universality the paper re-derives by transformation.","marker":"[22]"},{"why":"The temperature/curvature model whose soft-threshold 'temperature' interpretation motivates the Fermi-Dirac relaxation and the exponent formula.","marker":"[36]"},{"why":"Connects sigmoid linking probabilities to inverse temperature in graph ensembles, grounding the physical analogy for the parameter $\\beta$.","marker":"[45]"}],"fun_headline_variants":["Universal k^-2 law emerges from general thresholded hidden variables","Generalised threshold model locks scale-free exponent at 2","Softening threshold breaks universal k^-2, tunes exponent","New proof: thresholding hidden variables gives universal k^-2"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The universal $k^{-2}$ result assumes the transformed integral $L_\\Delta(\\tilde f)$ is finite and the size-independence condition $L_\\Delta(\\tilde f)=1/N$ can be met; the tunable-exponent result further assumes that the nodes dominating the power-law tail have hidden variables satisfying $H(x)\\ll\\Delta$, which the paper notes is only a good approximation for $\\beta\\in(0,1)$ and degrades as $\\beta\\to 0$.","fun_headline_variants_meta":{"raw":{"variants":["Universal k^-2 law emerges from general thresholded hidden variables","Generalised threshold model locks scale-free exponent at 2","Softening threshold breaks universal k^-2, tunes exponent","New proof: thresholding hidden variables gives universal k^-2"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000624,"raw_usage":{"total_tokens":2920,"prompt_tokens":1004,"completion_tokens":1916,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":620,"completion_tokens_details":{"reasoning_tokens":1847}},"tokens_in":620,"tokens_out":1916,"duration_ms":13844,"temperature":1.0,"reasoning_tokens":1847,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:02:18.777168+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the exponential-like model with $\\rho(x)=3x^2e^{-x^3}$, the soft link $f_{\\beta,\\Delta}=1/(1+e^{-\\beta(H(x)+H(y)-\\Delta)})$, and $N=10^5$ for $\\beta=0.3$ with $\\Delta=(1/\\beta)\\ln N$. Measure the degree exponent over the top three decades of $k$ and compare it with $1+1/\\beta\\approx 4.33$; if the effective exponent varies with the fitting window or the tail deviates from a pure power law, the approximation in Eq. (34) fails. For the hard-threshold claim, pick an $f$ with divergent $L_\\Delta(\\tilde f)$, for instance $\\tilde f(z)=e^{z/2}$ above the cutoff, and check that the degree distribution is no longer $k^{-2}$.","supporting_citations":[{"cited_title":"& Muñoz, M","cited_arxiv_id":null,"evidence_quote":"Introduces the threshold hidden-variable model whose exponential-fitness plus step-function case is the baseline the paper generalises."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the continuous approximation $k(x)=N\\int f\\rho\\,dy$ and the degree-transformation formula $p(k)=\\rho(x)/k'(x)$ used throughout."},{"cited_title":"& Pastor-Satorras, R","cited_arxiv_id":null,"evidence_quote":"Establishes the analytic hidden-variable framework and the size-independence condition for correlated random networks."},{"cited_title":"& Konno, N","cited_arxiv_id":null,"evidence_quote":"Earlier demonstration that threshold graphs with intrinsic vertex weights show the $k^{-2}$ degree decay the paper extends."},{"cited_title":"& Miwa, H","cited_arxiv_id":null,"evidence_quote":"Provides extreme-value-theory analysis of universal power laws in threshold networks, the universality the paper re-derives by transformation."},{"cited_title":"& Boguñá, M","cited_arxiv_id":null,"evidence_quote":"The temperature/curvature model whose soft-threshold 'temperature' interpretation motivates the Fermi-Dirac relaxation and the exponent formula."},{"cited_title":"E., Fink, T","cited_arxiv_id":null,"evidence_quote":"Connects sigmoid linking probabilities to inverse temperature in graph ensembles, grounding the physical analogy for the parameter $\\beta$."}],"review_version":1}