Pith. sign in

REVIEW 3 major objections 4 minor 33 references

Global and local synchrony of coupled neurons in small-world networks

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

Pith's one-line read In ordinary small-world networks, the optimal rewiring probability for global synchrony disappears when every neuron keeps the same number of connections, indicating that the peak comes from degree heterogeneity, not topology.

desk verdict A useful degree-preserving rewiring control and a plausible two-assembly mechanism, but the paper is likely an unlabeled ~2003 manuscript and the degree-heterogeneity attribution is not fully isolated. read the letter →

arxiv 2411.16374 v1 pith:MQSSV2BJ submitted 2024-11-25 q-bio.NC cond-mat.dis-nn

classification q-bio.NCcond-mat.dis-nn
keywords small-worldnetworksneuronalsynchronyleakyintegrate-and-fireneuronspulse-coupledoscillatorsdegreeheterogeneitybalancedrewiringfeaturebindingcoherentinputs
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

Global synchrony of pulse-coupled neurons in small-world networks is usually reported to be best at an intermediate rewiring probability, but this paper argues that the peak is an artifact of the standard rewiring procedure, which gives different neurons different numbers of connections. Using a modified 'balanced' rewiring that keeps every vertex degree identical while preserving the characteristic path length and clustering coefficient, the intermediate peak vanishes: global synchrony simply improves as rewiring increases. The paper then shows that when two spatially separated groups of neurons receive coherent inputs, local clustering and long-range shortcuts cooperate, with intermediate rewiring reinforcing synchrony between the distant groups even though it is larger than the value traditionally associated with the small-world regime. The result matters because it separates genuine topological effects on synchronization from side effects of connection-count heterogeneity, and because it offers a mechanism by which remote neural assemblies with a common stimulus can bind without requiring spatial proximity.

What carries the argument

The load-bearing construction is balanced random rewiring: edges are removed as in the ordinary procedure, the removed edge endpoints are paired through a random permutation ($v_{i,1}$ connected to $v_{\rho(i),2}$), and rewiring is rejected if it creates multiple edges or self-loops. This keeps every vertex degree exactly $k$ for all rewiring probabilities while leaving $L(p)$ and $C(p)$ nearly identical to the unbalanced case, so differences in dynamics can be attributed to degree heterogeneity. The other central quantity is the synchrony measure $syn(S)$, the ratio of the fluctuation of the group-averaged potential to the average of individual fluctuations, computed for groups of size $n'$; it lets the paper distinguish local precision (small $n'$) from global synchrony ($n'=400$).

What would settle it

Compute higher-order network statistics such as degree assortativity, motif counts, or degree correlations for balanced and unbalanced rewiring at the same $p$; if these statistics differ substantially and a synchrony difference persists when degree heterogeneity is controlled by an independent method (for example, a configuration model with a fixed degree sequence), then the attribution of the peak to $k_v$ alone would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that, in pulse-coupled leaky integrate-and-fire networks, the degree distribution of the network is a dynamical variable: heterogeneous numbers of inputs per neuron ($k_v$) act as a desynchronizing factor independent of the small-world statistics $L(p)$ and $C(p)$. When the standard Watts-Strogatz rewiring is replaced by balanced rewiring that keeps $k_v = k$ for all neurons while leaving $L(p)$ and $C(p)$ almost unchanged, the previously observed enhancement of global synchrony at intermediate $p$ disappears; synchrony now increases monotonically with $p$. In addition, the paper establishes a cooperation regime: for two remote groups receiving coherent inputs, intermediate rewiring gives intergroup synchrony that exceeds both the small-$p$ regime (where long-distance communication is too slow) and the large-$p$ regime (where local connectivity is lost), and this enhancement is robust to constant biases versus correlated noise and to the numerical integration scheme. The authors interpret this as evidence that local clustering supplies precise within-group locking while global shortcuts enable between-group locking, and that both must cooperate for binding of distant coherent stimuli.

Load-bearing premise

The load-bearing premise is that matching the average path length and clustering coefficient between balanced and unbalanced rewiring is enough to control every topology-dependent feature, so that the remaining dynamical difference can be attributed solely to heterogeneous vertex degrees.

Editorial extensions

If this is right

  • For pulse-coupled neurons, the optimal rewiring probability for global synchrony reported in earlier small-world studies is not a topological effect; networks with homogeneous connection counts should show monotonic improvement of global synchrony with randomness.
  • Local clustering is the carrier of precise local synchrony: as $p$ increases and clustering falls, local synchrony degrades even while global synchrony improves.
  • Remote groups receiving coherent inputs synchronize best at intermediate $p$, where local edges within each group and shortcuts between groups coexist; the required $p$ is larger than the classical small-world regime but still intermediate.
  • The brain could switch between precise local coding and rough global coding by modulating effective rewiring through synaptic plasticity.
  • Binding of remote features does not require closeness of stimuli, homogeneous global inhibition, or rapidly changing synaptic strengths; cooperative local and global connections suffice when inputs are coherent.

Reading between the lines

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

  • A direct extension would test whether connection-count heterogeneity acts through input-sum variance: in balanced networks, summing spike inputs from exactly $k$ neighbors removes one source of stochasticity, so one could predict that networks with fixed $k$ but Poisson-distributed synaptic weights would reintroduce a desynchronizing effect.
  • Because balanced rewiring preserves only $L$ and $C$, higher-order statistics such as degree correlations and motif counts may still differ; attributing the entire dynamical difference to $k_v$ assumes these higher-order features do not matter for pulse-coupled dynamics.
  • The cooperation mechanism suggests a testable prediction for experiments: if cortical networks sit in the intermediate regime, selectively silencing long-range connections should abolish intergroup synchrony for spatially separated coherent stimuli while leaving intragroup synchrony intact.
  • The balanced-rewiring methodology could be applied to other dynamical neuron models, such as Hodgkin-Huxley or FitzHugh-Nagumo systems, to re-examine previously reported optimal rewiring probabilities.
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 studies pulse-coupled leaky integrate-and-fire neurons on Watts-Strogatz-type small-world networks. It distinguishes ordinary ('unbalanced') rewiring, which produces heterogeneous vertex degrees, from a modified 'balanced' rewiring that keeps the degree sequence regular while leaving L(p) and C(p) approximately unchanged. The authors report that the intermediate-rewiring synchrony peak seen in ordinary small-world networks disappears under balanced rewiring, and they conclude that the peak is caused by heterogeneity in connection counts rather than by small-world topology. They then examine two spatially separated groups receiving coherent inputs and report enhanced intergroup synchrony for intermediate p, which they attribute to cooperation between local clustering and global shortcuts. The paper closes with a discussion of two coding modes and implications for feature binding.

Significance. If the central attribution is correct, the paper makes a useful conceptual contribution: it separates the effect of degree heterogeneity from the small-world properties L(p) and C(p), and it proposes a concrete mechanism by which local and global connectivity cooperate to synchronize remote neuronal assemblies. The simulation protocol is clearly specified, the balanced rewiring construction is explicitly defined, and the inclusion of an improved-Euler robustness check is a genuine strength. The paper also contains an explicit limitation statement in Section 7.3, acknowledging that random rewiring discards topology-dependent information. However, the main inference from the balanced-rewiring comparison is not fully isolated from higher-order topological changes, and the quantitative evidence for the cooperation claim is weakened by the absence of error bars and statistical tests.

major comments (3)
  1. [Section 4, Figs. 1-2] The central claim that the intermediate-p synchrony peak in ordinary small-world networks is caused by heterogeneity in kv is not established, because the balanced rewiring procedure can change more than the degree variance. The construction preserves the degree sequence exactly, but it may also alter the distribution of edge lengths, the number of long-range edges connecting the two stimulated groups, degree assortativity, and higher-order motif counts. The paper only reports L(p) and C(p) in Fig. 1, and Section 5.1 explicitly notes that a small L(p) 'does not necessarily indicate sufficient global connections.' Since the two graph families are compared only through L(p) and C(p), the observed difference in Fig. 2 could be due to these unmeasured topological differences rather than to degree heterogeneity. A control using an alternative degree-preserving rewiring protocol, or a direct quantification of cross-group and within-group edge counts as a function of p, is needed to support the attribution.
  2. [Section 5.1, Fig. 3(a)] The cooperation claim rests on small differences between the syn(S1 ∪ S2) and syn(S1 ∪ S2 ∪ S) curves at intermediate p, but the reported values are means over 25 runs with no error bars, confidence intervals, or statistical tests. The text states that intergroup synchrony is 'significantly larger' than syn(S1 ∪ S2 ∪ S), yet no significance test is reported. Given that the peak in Fig. 2(a) is itself described as 'not so prominent,' the reader cannot assess whether the apparent enhancement at p ≈ 0.3 in Fig. 3(a) is robust to run-to-run variability. Reporting standard errors or a permutation test would make the central claim quantitatively testable.
  3. [Section 6, Fig. 4] The conclusion that the 'additional amount of syn' in the correlated-noise case is 'induced by the global coupling' is based on a visual subtraction of the curves for input (II) and input (III). No uncertainty is attached to either curve, and the difference appears to be of similar magnitude to the run-to-run variability visible in other figures. Without error bars or a statistical comparison, this subtraction does not provide quantitative support for the claim that the enhancement at intermediate p is purely topological. This point is load-bearing for the paper's strongest conclusion that cooperation of local and global connectivity reinforces distant-group synchrony.
minor comments (4)
  1. [Fig. 1] The claim that L(p) and C(p) are unaffected by balanced rewiring is based on visual overlap of the two curves; no error bars or numerical values are provided. Adding a table of mean values, or at least error bars, would make the invariance claim verifiable.
  2. [Section 5.1] The robustness check with the improved Euler algorithm is shown only for (k, ε) = (12, 0.05) in Fig. 3(a). The text states that the conclusion 'is also true for other numerical results in this paper,' but no corresponding data are shown. Either show the comparison for the other parameter sets or soften the claim.
  3. [Section 3] Equation (2) defines syn(S) as a ratio of standard deviations, and the text says it is normalized between 0 and 1. This is true under the implicit assumption that all xi are nonnegative and have positive variance, but the condition is not stated explicitly. A one-sentence clarification would prevent ambiguity.
  4. [Section 5.2] The Ornstein-Uhlenbeck process in Eq. (4) is written with Gaussian white noise ξi(t), but no diffusion constant or variance parameter is specified. Since the noise amplitude is a free parameter and the results may depend on it, the exact stochastic differential equation used in the simulations should be stated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's synchrony results are simulation outputs, not fitted or definitionally tied to their inputs.

full rationale

The paper's central claims are supported by direct numerical simulations of pulse-coupled leaky integrate-and-fire neurons, with all parameters fixed in advance from standard modeling practice rather than fitted to reproduce the reported synchrony patterns. The synchrony measure syn(S) in Eq. (2) is defined independently of the conclusions, and the graph families are generated by explicit rewiring algorithms: the unbalanced construction is the standard Watts-Strogatz procedure and the balanced construction is a degree-preserving permutation of the rewired stubs. The observation that the intermediate-p synchrony peak disappears under balanced rewiring is a measured simulation contrast, not a quantity that is equal to its own input by construction. The paper does not derive any result from a fitted parameter renamed as a prediction, and no uniqueness theorem or load-bearing premise is imported from the authors' prior work; self-citations [16,17] appear only in the discussion of feedforward dual coding and are contextual rather than foundational. The skeptical concern that balanced rewiring may alter higher-order topological statistics beyond L(p) and C(p) is a legitimate control limitation, but it is a threat to causal attribution, not circularity: the simulation output is not equivalent by definition to the degree-variance input. Hence no circular step meets the required evidentiary standard.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper makes no analytic derivation and introduces no hidden fitted constants; its conclusions rest on simulation parameters chosen from earlier literature and on the assumption that balanced rewiring controls topology. Free parameters are standard model constants, but the central qualitative result is only explored in a narrow parameter window.

free parameters (5)
  • synaptic coupling strength epsilon = 0.05, 0.073, 0.035
    Chosen to keep firing rates fixed when k changes in Section 5.1; the intermediate-p synchrony enhancement depends on being in this coupling regime.
  • leak rate and bias current = gamma=25/40 ms^-1, I0=28.5/38.5/0.044
    Chosen to place neurons in an oscillatory regime; different regimes may change synchrony behavior.
  • input heterogeneity Delta I = 0.08, 0.1
    Controls the desynchronizing influence of heterogeneous inputs; results may depend on this spread.
  • noise amplitude and OU correlation = SD=1e-3, corr=0.8
    Correlated noise inputs in Section 5.2 are generated with these values; the reported separation between syn(S1 union S2) and syn(S1 union S2 union S) depends on the correlation level.
  • time delay and refractory period = tau=0.5/1.2 ms, tau'=0.6 ms
    Synaptic delay and refractory period are fixed; pulse-coupled synchronization is known to depend on these timing parameters.
assumptions (5)
  • domain assumption Neural connectivity is approximated by undirected graphs.
    Section 2 states undirected graphs are used even though real neural networks are directed, to simplify analysis.
  • domain assumption Pulse-coupled LIF dynamics with delta spikes, reset, and refractory period capture synchronization-relevant behavior.
    Section 3 defines the model; conclusions are about this model class, not about biophysically detailed neurons.
  • domain assumption The Watts-Strogatz rewiring procedure and the balanced rewiring variant generate representative network families with controlled L and C.
    Section 4 relies on Fig. 1 to claim L(p) and C(p) are insensitive to the balanced modification.
  • standard math Euler integration with 0.05 ms step approximates the true dynamics.
    Section 3 uses Euler; Section 5.1 checks improved Euler for Fig. 3(a) and states results do not critically depend on integration method.
  • domain assumption The synchrony measure syn(S) (Eq. 2) is a valid ordering of synchrony across group sizes and rewiring probabilities.
    Section 3 defines syn and the paper compares group sizes; the measure normalizes a population waveform fluctuation by averaged single-neuron fluctuations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Global and local synchrony of coupled neurons in small-world networks." pith.science (2026). https://pith.science/paper/MQSSV2BJ

@misc{pith2026241116374,
  author       = {Pith},
  title        = {Pith review of: Global and local synchrony of coupled neurons in small-world networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MQSSV2BJ}},
  note         = {Machine review of arXiv:2411.16374}
}
read the original abstract

Synchronous firing of neurons is thought to play important functional roles such as feature binding and switching of cognitive states. Although synchronization has mainly been investigated using model neurons with simple connection topology so far, real neural networks have more complex structures. Here we examine behavior of pulse-coupled leaky integrate-and-fire neurons with various network structures. We first show that the dispersion of the number of connections for neurons influences dynamical behavior even if other major topological statistics are kept fixed. The rewiring probability parameter representing the randomness of networks bridges two spatially opposite frameworks: precise local synchrony and rough global synchrony. Finally, cooperation of the global connections and the local clustering property, which is prominent in small-world networks, reinforces synchrony of distant neuronal groups receiving coherent inputs.

Figures

Figures reproduced from arXiv: 2411.16374 by the authors.

Figure 1
Figure 1. (a) L(p) and (b) C(p) for n = 400 and k = 6. The graphs are generated by the unbalanced random rewiring (crosses) and by the balanced random rewiring (circles). 2 Small-world Networks The characteristic path length L and the clustering coefficient C are commonly used to quantify the distances between vertices and the clustering property, respectively [32, 33]. With a graph composed of n vertices, L is defined to be … view at source ↗
Figure 2
Figure 2. The synchrony measure syn with (a) the unbalanced random rewiring, and (b) the balanced random rewiring. The size of local groups of neurons within which syn is evaluated is denoted by n ′ . edge, or self-connection. Insusceptibility of L(p) and C(p) to the graph modification evident in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The synchrony measure syn(S1) (crosses), syn(S2) (circles), syn(S1 ∪ S2) (asterisks), and syn(S1 ∪ S2 ∪S) (squares) when two separated coherent stimuli are presented. (a), (b), and (c) are the results for bias inputs with (k, ǫ) = (12, 0.05), (8, 0.073), and (18, 0.035), respectively. (d) shows the results for correlated noise inputs with (k, ǫ) = (12, 0.05). The results are obtained by the normal Euler algorithm ex… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The synchrony measure syn(S1 ∪ S2) when (I) a globally coherent input (solid lines), (II) two separated coherent inputs (dashed lines), and (III) two separated incoherent inputs (dotted lines) are presented. The inputs are applied as (a) biases or (b) correlated noise.…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Cambridge University Press, Cambr idge

    Abeles M (1991) Corticonics. Cambridge University Press, Cambr idge

  2. [2]

    Phys Rev Lett 89(5): 054101

    Barahona M, Pecora LM (2002) Synchronization in small-world sys tems. Phys Rev Lett 89(5): 054101

  3. [3]

    Proc Natl Acad Sci USA 92: 3844–3848

    Ben-Yishai R, Lev Bar-Or R, Sompolinsky H (1995) Theory of orie ntation tuning in visual cortex. Proc Natl Acad Sci USA 92: 3844–3848

  4. [4]

    Callaway DS, Hopcroft JE, Kleinberg JM, Newman MEJ, Strogatz S H (2001) Are randomly grown graphs really random? Phys Rev E 64: 041902

  5. [5]

    Springer-Verlag, New York

    Domany E, van Hemmen JK, Schulten K (Eds.) (1994) Models of neu ral networks II. Springer-Verlag, New York

  6. [6]

    Publicationes Mathematicae 6: 290–297

    Erd¨ os P, R´ enyi A (1959) On random graphs. Publicationes Mathematicae 6: 290–297

  7. [7]

    Neural Netw 9( 8): 1303–1350

    Fujii H, Ito H, Aihara K, Ichinose N, Tsukada M (1996) Dynamical cell assembly hypothesis – theoretical possibility of spatio-temporal coding in the cortex. Neural Netw 9( 8): 1303–1350

  8. [8]

    Neural Comput 12: 1095–1139

    Golomb D, Hansel D (2000) The number of synaptic inputs and the synchrony of large, sparse neuronal networks. Neural Comput 12: 1095–1139

Show all 33 references
  1. [9]

    Nature 338: 334–337

    Gray CM, K¨ onig P, Engel AK, Singer W (1989) Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338: 334–337

  2. [10]

    Phys Rev E 62(4): 5565–5570 8

    Guardiola X, D ´ ıaz-Guilera A, Llas M, P´ erez CJ (2000) Synchron ization, diversity, and topology of integrate and fire oscillators. Phys Rev E 62(4): 5565–5570 8

  3. [11]

    Neural Comput 10: 467–483

    Hansel D, Mato G, Meunier C, Neltner L (1998) On numerical simu lations of integrate-and-fire neural networks. Neural Comput 10: 467–483

  4. [12]

    synchroniza- tion

    K¨ onig P, Schillen TB (1991) Stimulus-dependent assembly forma tion of oscillatory responses: I. synchroniza- tion. Neural Comput 3: 155–166

  5. [13]

    Phys Rev Lett 84(12): 2758–2761

    Lago-Fern´ andez LF, Huerta R, Corbacho F, Sig¨ uenza JA (2000) Fast response and temporal coherent oscilla- tions in small-world networks. Phys Rev Lett 84(12): 2758–2761

  6. [14]

    Neural Netw 14: 687– 696

    Lago-Fern´ andez LF, Corbacho FJ, Huerta R (2001) Connec tion topology dependence of synchronization of neural assemblies on class 1 and 2 excitability. Neural Netw 14: 687– 696

  7. [15]

    Mainen ZF, Sejnowski TJ (1995) Reliability of spike timing in neocor tical neurons Science 268: 1503–1506

  8. [16]

    Phys Rev Lett 88(24): 248101

    Masuda N, Aihara K (2002) Bridging rate coding and temporal sp ike coding by effect of noise. Phys Rev Lett 88(24): 248101

  9. [17]

    Neural Comput 15: 103–125

    Masuda N, Aihara K (2003) Duality of rate coding and temporal c oding in multilayered feedforward networks. Neural Comput 15: 103–125

  10. [18]

    Random Struct Algorithms 6: 161–180

    Molloy M, Reed B (1995) A critical point for random graphs with a g iven degree sequence. Random Struct Algorithms 6: 161–180

  11. [19]

    Proc Natl Acad Sci USA 87: 7200–7204

    Sompolinsky H, Golomb D, Kleinfeld D (1990) Global processing of v isual stimuli in a neural network of coupled oscillators. Proc Natl Acad Sci USA 87: 7200–7204

  12. [20]

    Phys Rev A 43(12): 6990–7011

    Sompolinsky H, Golomb D, Kleinfeld D (1991) Cooperative dynamics in visual processing. Phys Rev A 43(12): 6990–7011

  13. [21]

    Cerebra l Cortex 10: 127–141

    Sporns O, Tononi G, Edelman GM (2000) Theoretical Neuroana tomy: relating anatomical and functional connectivity in graphs and cortical connection matrices. Cerebra l Cortex 10: 127–141

  14. [22]

    Nature 404: 18 7–190

    Steinmetz PN, Roy A, Fitzgerald PJ, Hsiao SS, Johnson KO, Niebu r E (2000) Attention modulates synchronized neuronal firing in primate somatosensory cortex. Nature 404: 18 7–190

  15. [23]

    Phil Trans R Soc Lond B 355: 111–126

    Stephan KE., Hilgetag C-C, Burns GAPC, O’Neill MA, Young MP, K¨ o tter R (2000) Computational analysis of functional connectivity between areas of primate cerebral co rtex. Phil Trans R Soc Lond B 355: 111–126

  16. [24]

    Physica 81D: 148–176

    Terman D, Wang D (1995) Global competition and local cooperat ion in a network of neural oscillators. Physica 81D: 148–176

  17. [25]

    J Neurosci 22(5): 1956–1966

    van Rossum MCW, Turrigiano GG, Nelson SB (2002) Fast propaga tion of firing rates through layered networks of noisy neurons. J Neurosci 22(5): 1956–1966

  18. [26]

    Phys Rev E 54(5): 5522–5537

    van Vreeswijk C (1996) Partial synchronization in populations o f pulse-coupled oscillators. Phys Rev E 54(5): 5522–5537

  19. [27]

    Science 274: 1724–1726

    van Vreeswijk C, Sompolinsky H (1996) Chaos in neuronal netwo rks with balanced excitatory and inhibitory activity. Science 274: 1724–1726

  20. [28]

    Neural Comput 10: 1321–1371

    van Vreeswijk C, Sompolinsky H (1998) Chaotic balanced state in a model of cortical circuits. Neural Comput 10: 1321–1371

  21. [29]

    Biol Cybern 54: 29–40

    von der Malsburg Ch, Schneider W (1986) A neural cocktail-par ty processor. Biol Cybern 54: 29–40

  22. [30]

    IEEE Trans on Neural Netw 6(4): 941–948

    Wang D (1995) Emergent synchrony in locally coupled neural osc illations. IEEE Trans on Neural Netw 6(4): 941–948

  23. [31]

    Journal of Neurosci 26(20): 6402–6413

    Wang XJ, Buzs´ aki G (1996) Gamma oscillation by synaptic inhibitio n in a hippocampal interneuronal network model. Journal of Neurosci 26(20): 6402–6413

  24. [32]

    Nature 393: 440–442

    Watts DJ, Strogatz SH (1998) Collective dynamics of ‘small-world ’ networks. Nature 393: 440–442

  25. [33]

    Princeton University Press, Prin ceton 9

    Watts DJ (1999) Small worlds. Princeton University Press, Prin ceton 9

Pith tools

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