{"id":"6504c9ea-74a8-4e08-b20a-ff5cb908c23b","arxiv_id":"2411.09098","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Three scaling exponents measured in human brain fMRI activity follow two linear relations, derived from a mean-field model, echoing scaling relations near critical points.","lead":"Researchers applied a renormalization group method to resting-state fMRI scans of 714 people and found that three scaling exponents of brain activity are linearly related to one another. The result suggests the brain's large-scale dynamics may follow universal scaling rules analogous to those found near phase transitions in physics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The analytical derivation of eβ=(3−eα)/2 fails at Eq. (12): the product in Eq. (10) telescopes to E(K)=K/2, so the claimed E∼K^{3/2}/M2^{1/2} and the resulting scaling relation do not follow.","rationale":"The paper's empirical observation of linear interdependencies across 714 subjects is interesting, and the relation eϵ=eα−1 does follow from the exact identity λ1=M2/K. However, the analytical claim for eβ is central to the paper's framing as 'scaling relations near critical points.' My check shows that Eq. (10) collapses to K/2, so Eq. (12) is not a consequence. This is a mathematical error, not merely an unvalidated biological assumption. It means the mean-field derivation of eβ=(3−eα)/2 is invalid as written. The reader's weakest assumption (F≈E) is real but downstream: even granting F≈E, the E(K) used does not scale as K^{(3−eα)/2}. I therefore keep the conditional verdict, but for a stronger reason: the analytical derivation must be corrected or removed, and the eβ relation should be treated as an empirical fit until a valid derivation is supplied. I also note the eϵ fit in Eq. (5) appears internally inconsistent with the third fit (which implies eϵ≈1.04eα−1.0, not +0.98), further supporting a conditional rather than accept verdict.","tokens_in":7457,"tokens_out":13547,"duration_ms":142329,"concrete_test":"Symbolically simplify Eq. (10): the product over n collapses, yielding E(K)=K/2 exactly. Then plug K=4, v=1, c=0.5 into Eq. (10) and Eq. (11): Eq. (10) gives 2, whereas Eq. (12)'s expression K^{3/2}/M2^{1/2} gives 8/√10≈2.53. More decisively, derive the scaling of F(K)=−log Z directly from the Gaussian partition function Z=(2π)^{K/2}(det C)^{1/2} with det C=(v−c)^{K−1}(v+(K−1)c); for fixed v,c this free energy grows linearly in K, so eβ=1, contradicting Eq. (13) for any eα<3. This settles whether the claimed analytical relation follows from the mean-field model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing concern is the derivation of eβ=(3−eα)/2, specifically the step from Eqs. (10)–(12). For any v, c, K the product in Eq. (10) telescopes exactly: ∏_{n=0}^{K−2} ((v+c+nc)/(v+nc))^{1/2} = ((v+(K−1)c)/v)^{1/2}. Multiplying by the prefactor v^{1/2}K/[2(v+(K−1)c)^{1/2}] gives E(K)=K/2 exactly. Consequently Eq. (10) does not depend on c or on M2, and it cannot imply E∼K^{3/2}/M2^{1/2}. Indeed, with Eq. (11), M2=vK+cK(K−1)∼cK^2, so the right-hand side of Eq. (12) scales as K^{3/2}/K = K^{1/2}, not E∼K. The mean-energy scaling E∼K^{3/2}/M2^{1/2} is therefore algebraically false. Taking the correct E(K)=K/2 would give eβ=1 for all eα in this mean-field model, not eβ=(3−eα)/2. The subsequent remark that F≈E (between Eqs. (12) and (13)) is beside the point: the claimed E-scaling is already unsupported. If E(K) was intended to denote some other quantity, that quantity and its relation to F(K)=−log P0 must be redefined and recomputed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies a phenomenological renormalization group to resting-state fMRI time series from 714 HCP subjects. By recursively coarse-graining the data, the authors compute scaling exponents eα, eβ, and eϵ for the variance, log probability of silence, and largest covariance eigenvalue. They report strong linear interdependencies among these exponents, propose analytical mean-field derivations yielding eβ = (3 − eα)/2 and eϵ = eα − 1, and further correlate eα with gray matter volume and cognitive performance. The central claim is that these scaling relations are intrinsic to brain organization and resemble thermodynamic scaling relations near critical points.","tokens_in":7904,"tokens_out":9148,"duration_ms":89458,"significance":"If the empirical interdependencies and their analytical derivation were correct, the paper would provide a striking example of scaling relations in a complex biological system, analogous to critical-point scaling in statistical physics. The use of a large public dataset and the reported high R² values make the empirical observation potentially valuable. The correlations with anatomical and behavioral traits are also of interest. However, the theoretical derivation is the main load-bearing contribution, and it is mathematically flawed, as detailed below.","major_comments":[{"comment":"The product in Eq. (10) telescopes exactly: ∏_{n=0}^{K−2} ((v+c+nc)/(v+nc))^{1/2} = ((v+(K−1)c)/v)^{1/2}, which cancels the prefactor and yields E(K) = K/2. The claim that \"for large K, the fractions inside the product tend to unity\" is misleading because the product as a whole does not tend to 1; it grows with K. Consequently, Eq. (12), E ∼ K^{3/2}/M2^{1/2}, is algebraically false, and the derived scaling relation eβ = (3 − eα)/2 does not follow. For the stated Gaussian model, the correct mean energy is E = K/2, which would give eβ = 1 for all eα, not the observed variation. This invalidates the central analytical derivation.","section":"Mean-field derivation, Eq. (10)–(12)"},{"comment":"The three reported regression equations are internally inconsistent. Combining eϵ = 1.02 eα + 0.98 and eβ = −0.52 eα + 1.56 gives eβ ≈ −0.51 eϵ + 2.06, not the reported eβ = −0.50 eϵ + 1.06. This suggests a likely typo, perhaps the second equation should read eα = 1.02 eϵ + 0.98, but as printed the equations cannot all hold simultaneously. The text also states that the empirical relations accommodate the surrogate data (eα ≈ 1.17, eϵ ≈ 0.20) and the independent limit (eα = 1, eϵ = 0), yet the printed eϵ = 1.02 eα + 0.98 gives eϵ ≈ 2.17 and 2.00 for those cases, contrary to the claim.","section":"Eq. (5)"},{"comment":"The argument that the entropy contribution to the free energy is negligible (F ≈ E) is beside the point because the claimed scaling of E is itself unsupported by Eq. (10). Furthermore, for the Gaussian model in Eq. (7), the free energy is F = −log Z = (1/2) log det C + const ≈ (K/2) log(v − c) + (1/2) log K, which scales linearly with K, giving eβ = 1, not eβ = (3 − eα)/2. The relation between the mean energy defined in Eq. (10) and the free energy used in Eq. (3) therefore needs to be clarified and recomputed.","section":"Mean-field derivation, between Eqs. (12) and (13)"}],"minor_comments":[{"comment":"The binarization threshold (z-score > 2) is a free parameter; the robustness of the exponents and of the linear relations to this threshold is not reported.","section":"Methods, binarization"},{"comment":"The caption mentions \"Csize = 64\" but the text says \"the last iterative step, Csize = 64\"; please ensure consistent notation for cluster size.","section":"Fig. 1"},{"comment":"It would be helpful to overlay the theoretical lines eβ = (3 − eα)/2 and eϵ = eα − 1 on the empirical scatter plots to make the claimed agreement visually assessable.","section":"Fig. 2(a)"},{"comment":"The sentence \"a fraction of 10% of the interval of Cij-values concentrate about 60% of all covariances\" is unclear; please specify the interval and define the percentage precisely.","section":"Statistical analysis"}],"recommendation":"major_revision","confidential_remarks":"The paper presents an intriguing empirical observation, but the analytical derivation contains an elementary algebraic error that invalidates the central theoretical claim. The inconsistency in Eq. (5) is likely a typo but should be corrected. Given that the empirical correlations with anatomy and behavior are plausibly real, the manuscript could be revised to present the empirical scaling relations as the main result, with a corrected or substantially weakened theoretical interpretation. As written, the load-bearing theory is not sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe empirical claim in this paper is worth a look: across 714 HCP subjects, the PRG scaling exponents eα, eβ, eϵ for rs-fMRI time series fall roughly on a line. To my knowledge, no one has reported across-subject interdependencies of these exponents before; previous PRG studies only gave population averages. That part is a real, if modest, contribution.\n\nThe surrogate-data control and the session-to-session stability are nice touches. The relation eϵ = eα - 1 is trivial to derive (λ1 = M2/K) and it holds up.\n\nThe trouble is the analytic derivation of eβ = (3 - eα)/2. Eq. (10) as written telescopes exactly: the product over n from 0 to K-2 of (v+(n+1)c)/(v+nc) is (v+(K-1)c)/v, and after multiplying by the prefactor, E(K) = K/2. So the claimed scaling E~K^{3/2}/M2^{1/2} does not follow; with E=K/2 you'd get eβ=1 for all eα. The step between Eq. (12) and Eq. (13) is therefore unsupported, and the F≈E approximation is beside the point. This is not a minor typo; it's the load-bearing derivation for one of the two reported scaling relations.\n\nThere's also a consistency issue in Eq. (5). The three bivariate regressions are not projections of a single line; combining the first two gives eβ = -0.51 eϵ + 2.06, not the reported -0.50 eϵ + 1.06. That suggests the linear collapse is approximate at best, and the fits should be presented with confidence intervals.\n\nThe paper gives no error bars on the individual exponents and no code, which makes it hard to check whether the linear trend survives other binarization thresholds (z=2 is just one choice). The gray matter and PMAT correlations are weak (ρ≈0.3, 0.16) but with 714 subjects they're significant; I'd treat those as secondary.\n\nBottom line: the empirical observation of linear interdependencies is plausible and new, but the theoretical explanation as written is wrong. This needs a major revision, not a minor patch. If the authors can either fix the algebra or present the eβ relation as purely empirical, the paper becomes interesting again.\n\nI'd send it to review—the empirical finding deserves scrutiny—but I'd also tell the referee to focus hard on Eqs. (10)-(13) and on the consistency of Eq. (5).","headline":"New empirical finding on PRG exponent correlations, but the analytic derivation has a load-bearing algebraic error.","tokens_in":8353,"tokens_out":3759,"would_cite":false,"duration_ms":36192,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Brain-activity scaling exponents collapse to a single line: one exponent fixes the other two.","keywords":["phenomenological renormalization group","resting-state fMRI","scaling exponents","mean-field model","critical phenomena","brain activity","coarse-graining","universality"],"falsifier":"Estimate the cluster free energy directly by measuring the probability $P_0$ of complete silence and independently measure the mean energy from the fitted pairwise model; if the difference $F-E$ (the entropy contribution) is not small compared with $E$ at the cluster sizes used in the fits, the predicted relation $e_\\beta=(3-e_\\alpha)/2$ should break down, and subjects or conditions with large entropy corrections should fall off the empirical line.","tokens_in":7274,"feed_emoji":"🧠","tokens_out":14552,"duration_ms":119436,"temperature":0.7,"pith_summary":"This paper claims that the scaling of coarse-grained resting-state brain activity is governed by one exponent, not three independent ones. Applying the phenomenological renormalization group to binarized fMRI time series from 714 subjects, the authors measure exponents $e_\\alpha$ for the variance, $e_\\beta$ for the log probability of silence, and $e_\\epsilon$ for the largest covariance eigenvalue as clusters grow. Across subjects the exponents fall on a single line in exponent space, with empirical fits $e_\\beta = -0.52e_\\alpha + 1.56$ and $e_\\epsilon = 1.02e_\\alpha + 0.98$. A mean-field calculation with a homogeneous covariance matrix derives the simpler relations $e_\\beta = (3-e_\\alpha)/2$ and $e_\\epsilon = e_\\alpha - 1$, which match the data. If this holds, measuring one exponent predicts the other two, and the brain joins a pattern of interdependent scaling familiar from critical phenomena.","feed_headline":"One exponent fixes two others in brain activity","feed_subtitle":"Variance, silence, and eigenvalue scaling obey two linear relations across 714 brains.","key_machinery":"The central object is the phenomenological renormalization group (PRG), a coarse-graining procedure in which the two most correlated regions of the brain are merged at each step, producing clusters of size $K$ whose statistical observables scale as power laws. The three observables are the variance $M_2(K)\\sim K^{e_\\alpha}$, the free energy $F(K)=-\\log P_0(K)\\sim K^{e_\\beta}$ where $P_0$ is the probability that a whole cluster is silent, and the largest covariance eigenvalue $\\lambda_1(K)\\sim K^{e_\\epsilon}$. The argument that links these exponents is a mean-field Gaussian cluster model with covariance matrix $C=(v-c)I_K + c\\,1_K1_K^T$; from it the identities $M_2=vK+cK(K-1)$, $\\lambda_1=M_2/K$, and $E\\sim K^{3/2}/M_2^{1/2}$ follow, and with the approximation $F\\approx E$ they collapse to the exponent relations.","core_discovery":"On the paper's own terms, the discovery is that the exponents $e_\\alpha$, $e_\\beta$, and $e_\\epsilon$ of resting-state fMRI activity are not scattered independently across subjects; they collapse onto a single curve described by the linear relations in Eq. (5). The authors account for these relations analytically in a mean-field picture in which each coarse-grained cluster is a multivariate Gaussian with uniform variance $v$ and uniform covariance $c$. In that model the variance of a cluster sum scales as $M_2(K)=vK+cK(K-1)$, the largest covariance eigenvalue is $\\lambda_1(K)=M_2(K)/K$, and the mean energy scales as $E(K)\\sim K^{3/2}/M_2(K)^{1/2}$; using the free energy $F=-\\log P_0(K)$ in place of $E$ gives $e_\\beta=(3-e_\\alpha)/2$, while $\\lambda_1\\sim K^{e_\\alpha-1}$ gives $e_\\epsilon=e_\\alpha-1$. The empirical fits in Fig. 2(a) lie close to these lines, surrogate phase-shuffled data fall near the trivial Gaussian values, and $e_\\alpha$ is correlated with gray matter volume and with cognitive-test performance.","pith_inferences":["The paper does not test whether the derived lines hold away from resting-state conditions; a direct extension would be to apply the PRG to task-evoked fMRI or to synthetic Curie-Weiss data with controlled coupling strength, checking whether deviations from the lines track distance from a critical regime.","Because the mean-field derivation strips out heterogeneity in variances and covariances, regional or clinical deviations from the fitted exponents could be read as a signature of heterogeneous functional connectivity, an interpretive step the authors do not take.","If these interdependencies are generic to multiscale time series, then an experiment that can measure only one scaling observable could infer the other two, allowing scaling analyses in data regimes where silence probabilities or eigenvalue spectra are poorly sampled."],"forward_implications":["Measuring $e_\\alpha$ alone predicts $e_\\beta$ and $e_\\epsilon$ for a subject, so future studies can report a single scaling exponent instead of three.","The relations automatically contain the Gaussian/independent limit ($e_\\alpha=1$, $e_\\beta=1$, $e_\\epsilon=0$) and approximately accommodate the surrogate-data exponents, so the mean-field lines serve as a baseline separating nontrivial from trivial scaling.","Because $e_\\alpha$ is correlated with gray matter volume and cognitive performance, the exponents and their interdependencies become a candidate subject-level descriptor of brain organization.","The same coarse-graining analysis applied to other multiscale time series should, if the claim generalizes, reveal similar linear relations among variance, silence, and eigenvalue exponents."],"supporting_citations":[{"why":"Supplies the phenomenological renormalization-group coarse-graining algorithm and defines the three scaling observables.","marker":"[10]"},{"why":"Provides the large resting-state fMRI dataset of 714 subjects from which all empirical exponents are computed.","marker":"[28]"},{"why":"Defines the 1014-region parcellation that fixes the set of raw variables coarse-grained in the PRG.","marker":"[29]"},{"why":"Documents the near-Gaussian activity statistics in resting-state fMRI that justify the multivariate-normal cluster model.","marker":"[30]"},{"why":"The earlier neural-network PRG result that entropy contributes appreciably less than mean energy, the step behind the e_beta relation.","marker":"[33]"},{"why":"Establishes pairwise maximum-entropy models as the basis for the Ising-like cluster distribution in Eq. (6).","marker":"[34]"},{"why":"Supplies the multivariate central limit theorem used to justify convergence to the multivariate normal distribution.","marker":"[36]"},{"why":"Prior PRG application using the same upper-threshold binarization of BOLD activity and reporting trivial surrogate-data exponents.","marker":"[13]"},{"why":"Provides the Fourier-phase-shuffle surrogate procedure used as the Gaussian baseline.","marker":"[32]"}],"fun_headline_variants":["Brain scaling exponents collapse to a line","Two equations determine all three brain exponents","Brain exponents obey two linear rules","A single curve unites brain scaling exponents","Brain activity exponents are tightly coupled"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that, for the resting-state fMRI clusters, the free energy can be replaced by the mean energy alone — that is, the entropy term in $F=-\\log P_0$ is negligible — an approximation the paper takes from earlier neural-network PRG studies rather than verifying directly on these data.","fun_headline_variants_meta":{"raw":{"variants":["Brain scaling exponents collapse to a line","Two equations determine all three brain exponents","Brain exponents obey two linear rules","A single curve unites brain scaling exponents","Brain activity exponents are tightly coupled"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001214,"raw_usage":{"total_tokens":4963,"prompt_tokens":878,"completion_tokens":4085,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":494,"completion_tokens_details":{"reasoning_tokens":4025}},"tokens_in":494,"tokens_out":4085,"duration_ms":29723,"temperature":1.0,"reasoning_tokens":4025,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:02:50.812438+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate the cluster free energy directly by measuring the probability $P_0$ of complete silence and independently measure the mean energy from the fitted pairwise model; if the difference $F-E$ (the entropy contribution) is not small compared with $E$ at the cluster sizes used in the fits, the predicted relation $e_\\beta=(3-e_\\alpha)/2$ should break down, and subjects or conditions with large entropy corrections should fall off the empirical line.","supporting_citations":[{"cited_title":"Meshulam, J","cited_arxiv_id":null,"evidence_quote":"Supplies the phenomenological renormalization-group coarse-graining algorithm and defines the three scaling observables."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the large resting-state fMRI dataset of 714 subjects from which all empirical exponents are computed."},{"cited_title":"Schaefer, R","cited_arxiv_id":null,"evidence_quote":"Defines the 1014-region parcellation that fixes the set of raw variables coarse-grained in the PRG."},{"cited_title":"Hindriks, T","cited_arxiv_id":null,"evidence_quote":"Documents the near-Gaussian activity statistics in resting-state fMRI that justify the multivariate-normal cluster model."},{"cited_title":"Tkaˇ cik, T","cited_arxiv_id":null,"evidence_quote":"The earlier neural-network PRG result that entropy contributes appreciably less than mean energy, the step behind the e_beta relation."},{"cited_title":"Schneidman, M","cited_arxiv_id":null,"evidence_quote":"Establishes pairwise maximum-entropy models as the basis for the Ising-like cluster distribution in Eq. (6)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the multivariate central limit theorem used to justify convergence to the multivariate normal distribution."},{"cited_title":"Ponce-Alvarez, M","cited_arxiv_id":null,"evidence_quote":"Prior PRG application using the same upper-threshold binarization of BOLD activity and reporting trivial surrogate-data exponents."},{"cited_title":"Schreiber and A","cited_arxiv_id":null,"evidence_quote":"Provides the Fourier-phase-shuffle surrogate procedure used as the Gaussian baseline."}],"review_version":1}