Pith. sign in

REVIEW 3 major objections 4 minor 39 references

Interdependent scaling exponents in the human brain

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Brain-activity scaling exponents collapse to a single line: one exponent fixes the other two.

desk verdict New empirical finding on PRG exponent correlations, but the analytic derivation has a load-bearing algebraic error. read the letter →

arxiv 2411.09098 v1 pith:L4JROYSW submitted 2024-11-14 cond-mat.dis-nn cond-mat.stat-mechnlin.AOphysics.data-anq-bio.NC

classification cond-mat.dis-nncond-mat.stat-mechnlin.AOphysics.data-anq-bio.NC
keywords phenomenologicalrenormalizationgroupresting-statefMRIscalingexponentsmean-fieldmodelcriticalphenomenabrainactivitycoarse-graininguniversality
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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.

Load-bearing premise

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.

Editorial extensions

If this is right

  • 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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • 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.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

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.

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 (3)
  1. [Mean-field derivation, Eq. (10)–(12)] 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.
  2. [Eq. (5)] 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.
  3. [Mean-field derivation, between Eqs. (12) and (13)] 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.
minor comments (4)
  1. [Methods, binarization] 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.
  2. [Fig. 1] The caption mentions "Csize = 64" but the text says "the last iterative step, Csize = 64"; please ensure consistent notation for cluster size.
  3. [Fig. 2(a)] 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.
  4. [Statistical analysis] 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.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical scaling relations are direct data measurements and the model derivation, though algebraically flawed, does not reduce to its own inputs.

full rationale

The central empirical claim—that the PRG exponents eα, eβ, eϵ extracted from rs-fMRI data satisfy the linear relations of Eq. (5)—is a direct data-analysis result, not an output of the fitted model. The exponents are measured independently through Eqs. (2)–(4) and are validated against surrogate and independent-variable limits. The analytic mean-field derivation does not fit eβ or eϵ to the empirical values: it assumes a homogeneous-covariance Gaussian model, derives the identity λ1(K)=M2(K)/K (Eq. 14), and then obtains eϵ=eα−1 as a genuine model consequence. The attempted derivation of eβ=(3−eα)/2 from E(K) is not circular, but it is mathematically unsupported: the product in Eq. (10) telescopes exactly to ((v+(K−1)c)/v)^{1/2}, so E(K)=K/2 identically, and the asserted E∼K^{3/2}/M2^{1/2}∼K^{(3−eα)/2} does not follow. That is a correctness problem, not a circularity, and it does not make the empirical relation equivalent to the model's inputs. The self-citations ([14], [30], [39]) are background, data-source, or preprocessing references and are not load-bearing for the scaling-relation claim. The paper also explicitly acknowledges the homogeneous-covariance limitation. Therefore there is no significant circularity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central relations depend on four assumptions: Gaussian convergence of coarse-grained variables, homogeneous mean-field covariance, a negligible entropy contribution to the free energy, and a simplification of the product in Eq. (10). The first is standard mathematics; the other three are model choices particular to this paper, two of which are not fully validated.

free parameters (1)
  • binarization threshold (z-score) = 2 (standard deviations above the mean)
    Chosen to convert BOLD signals to binary variables; no robustness analysis across thresholds is provided, and exponent values likely depend on this choice.
assumptions (4)
  • standard math The coarse-grained activity distribution converges to a multivariate Gaussian (multivariate central limit theorem applies).
    Invoked around Eq. (7); a standard theorem, but its applicability to the PRG-merging procedure is an assumption.
  • ad hoc to paper Covariance matrix of cluster variables has the homogeneous form Cii=v, Cij=c for i≠j.
    Mean-field Curie-Weiss-style assumption used in Eq. (8); explicitly acknowledged as a limitation in the concluding paragraph.
  • ad hoc to paper Entropy contribution to the free energy is negligible relative to the mean energy (F ≈ E).
    Used to obtain Eq. (13); justified by a citation to Ref. [33] for neural data, not validated for rs-fMRI.
  • ad hoc to paper For large K, the product terms in Eq. (10) tend to unity.
    Used to simplify E(K) to K^{3/2}/M2^{1/2}; the product actually grows as a power law if c>0, so this is an uncontrolled approximation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interdependent scaling exponents in the human brain." pith.science (2026). https://pith.science/paper/L4JROYSW

@misc{pith2026241109098,
  author       = {Pith},
  title        = {Pith review of: Interdependent scaling exponents in the human brain},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L4JROYSW}},
  note         = {Machine review of arXiv:2411.09098}
}
read the original abstract

We apply the phenomenological renormalization group to resting-state fMRI time series of brain activity in a large population. By recursively coarse-graining the data, we compute scaling exponents for the series variance, log probability of silence, and largest covariance eigenvalue. The exponents clearly exhibit linear interdependencies, which we derive analytically in a mean-field approach. We find a significant correlation of exponent values with the gray matter volume and cognitive performance. Akin to scaling relations near critical points in thermodynamics, our findings suggest scaling interdependencies are intrinsic to brain organization and may also exist in other complex systems.

Figures

Figures reproduced from arXiv: 2411.09098 by the authors.

Figure 1
Figure 1. FIG. 1. (a)-(d) Illustrative scheme of the coarse-graining method. (a)-(b) We first obtain [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. (a) Linear relations between scaling exponents ( [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 30 canonical work pages

  1. [1]

    Domb, The Critical Point: A Historical Introduc- tion to the Modern Theory of Critical Phenomena (CRC Press, London, 1996)

    C. Domb, The Critical Point: A Historical Introduc- tion to the Modern Theory of Critical Phenomena (CRC Press, London, 1996)

  2. [2]

    K. G. Wilson, Rev. Mod. Phys. 47, 773 (1975)

  3. [3]

    L. P. Kadanoff, J. Stat. Phys. 137, 777 (2009)

  4. [4]

    Barab´ asi and R

    A.-L. Barab´ asi and R. Albert, Science286, 509 (1999)

  5. [5]

    J. P. Sethna, K. A. Dahmen, and C. R. Myers, Nature 410, 242 (2001)

  6. [6]

    Touboul and A

    J. Touboul and A. Destexhe, Phys. Rev. E 95, 012413 (2017)

  7. [7]

    M. C. Morrell, A. Sederberg, and I. Nemenman, Phys. Rev. Let. 126, 118302 (2021)

  8. [8]

    J. M. Beggs and N. Timme, Front. Physiol. 3, 163 (2012)

Show all 39 references
  1. [9]

    G. T. Cantwell, Y. Liu, B. F. Maier, A. C. Schwarze, C. A. Serv´ an, J. Snyder, and G. St-Onge, Phys. Rev. E 101, 062302 (2020)

  2. [10]

    Meshulam, J

    L. Meshulam, J. L. Gauthier, C. D. Brody, D. W. Tank, and W. Bialek, Phys. Rev. Lett. 123, 178103 (2019)

  3. [11]

    Nicoletti, S

    G. Nicoletti, S. Suweis, and A. Maritan, Phys. Rev. Res. 2, 023144 (2020)

  4. [12]

    Morales, S

    G. Morales, S. Di Santo, and M. Mu˜ noz, Proc. Natl. Acad. Sci. USA 120, e2208998120 (2023)

  5. [13]

    Ponce-Alvarez, M

    A. Ponce-Alvarez, M. L. Kringelbach, and G. Deco, Commun. Biol. 6, 627 (2023)

  6. [14]

    D. M. Castro, T. Feliciano, N. A. P. de Vasconcelos, C. Soares-Cunha, B. Coimbra, A. J. Rodrigues, P. V. Carelli, and M. Copelli, PRX Life 2, 023008 (2024)

  7. [15]

    B. R. Munn, E. M¨ uller, I. Favre-Bulle, E. Scott, M. Breakspear, and J. M. Shine, bioRxiv , 2024.06.22.600219 (2024)

  8. [16]

    Meshulam and W

    L. Meshulam and W. Bialek, arXiv:2409.00412 [cond- mat.dis-nn] (2024)

  9. [17]

    C. Song, S. Havlin, and H. A. Makse, Nature 433, 392 (2005)

  10. [18]

    C. Song, S. Havlin, and H. A. Makse, Nat. Phys. 2, 275 (2006)

  11. [19]

    Radicchi, A

    F. Radicchi, A. Barrat, S. Fortunato, and J. J. Ramasco, Phys. Rev. E 79, 026104 (2009)

  12. [20]

    Garuccio, M

    E. Garuccio, M. Lalli, and D. Garlaschelli, Phys. Rev. Res. 5, 043101 (2023)

  13. [21]

    M. ´A. Serrano, M. Bogun´ a, and A. Vespignani, Proc. Natl. Acad. Sci. USA 106, 6483 (2009)

  14. [22]

    Garc ´ ıa-P´ erez, M

    G. Garc ´ ıa-P´ erez, M. Bogu˜ n´ a, and M.´A. Serrano, Nat. Phys. 14, 583 (2018)

  15. [23]

    Zheng, A

    M. Zheng, A. Allard, P. Hagmann, Y. Alem´ an-G´ omez, and M. ´A. Serrano, Proc. Natl. Acad. Sci. USA 117, 20244 (2020)

  16. [24]

    Barjuan, M

    L. Barjuan, M. Zheng, and M. Serrano, arXiv:2410.06739 [q-bio.NC] (2024)

  17. [25]

    Villegas, T

    P. Villegas, T. Gili, G. Caldarelli, and A. Gabrielli, Nat. Phys. 19, 445 (2023)

  18. [26]

    M. C. Loures, A. A. Piovesana, and J. A. Brum, arXiv:2302.07093 [cond-mat.dis-nn] (2023)

  19. [27]

    Nurisso, M

    M. Nurisso, M. Morandini, M. Lucas, F. Vaccarino, T. Gili, and G. Petri, arXiv:2401.11298 [cond-mat.stat- mech] (2024)

  20. [28]

    D. C. Van Essen, K. Ugurbil, E. Auerbach, D. Barch, T. E. J. Behrens, R. Bucholz, A. Chang, L. Chen, M. Cor- betta, S. W. Curtiss, et al., Neuroimage 62, 2222 (2012)

  21. [29]

    Schaefer, R

    A. Schaefer, R. Kong, E. M. Gordon, T. O. Laumann, X. Zuo, A. J. Holmes, S. B. Eickhoff, and B. T. Yeo, Cereb. Cortex 28, 3095 (2018)

  22. [30]

    Hindriks, T

    R. Hindriks, T. A. A. Broeders, M. M. Schoonheim, L. Douw, F. Santos, W. van Wieringen, and P. K. Tewarie, Hum. Brain Mapp. 45, e26663 (2024)

  23. [31]

    Tagliazucchi, P

    E. Tagliazucchi, P. Balenzuela, D. Fraiman, and D. R. Chialvo, Front. Physiol. 3, 15 (2012)

  24. [32]

    Schreiber and A

    T. Schreiber and A. Schmitz, Physica D 142, 346 (2000)

  25. [33]

    Tkaˇ cik, T

    G. Tkaˇ cik, T. Mora, O. Marre, D. Amodei, S. E. Palmer, M. J. Berry II, and W. Bialek, Proc. Natl. Acad. Sci. USA 112, 11508 (2015)

  26. [34]

    Schneidman, M

    E. Schneidman, M. J. Berry, R. Segev, and W. Bialek, Nature 440, 1007 (2006)

  27. [35]

    Meshulam, J

    L. Meshulam, J. L. Gauthier, C. D. Brody, D. W. Tank, and W. Bialek, arXiv:1812.11904 [physics.bio-ph] (2018)

  28. [36]

    T. W. Anderson, An Introduction to Multivariate Statis- tical Analysis (Wiley, New York, 2003)

  29. [37]

    Cocchi, L

    L. Cocchi, L. L. Gollo, A. Zalesky, and M. Breakspear, Prog. Neurobiol. 158, 132 (2017)

  30. [38]

    Zimmern, Front

    V. Zimmern, Front. Neural Circuits 14, 565335 (2020)

  31. [39]

    Human connec- tome project rfmri repository,

    F. Tijhuis, M. Schepers, E. Centeno, B. d. A. Maciel, L. Douw, and F. Nobrega Santos, “Human connec- tome project rfmri repository,” https://github.com/ floristijhuis/HCP-rfMRI-repository (2021)

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.