Open core of the renormalization growth theory of fractal complex networks: scale-invariant hub–hub correlation diagnostics, reliable exponent estimation near the small-world limit, and the robustness-selection hypothesis

Song, Havlin and Makse (Nature Physics 2, 275–281, 2006) explained the fractal (self-similar) architecture of many real modular networks by a renormalization-driven growth model in which hubs effectively repel each other across all length scales, and conjectured that fractality is evolutionarily favored because it raises robustness to targeted hub attacks. The paper itself declared three sharp unresolved issues: standard scalar correlation measures such as the Pearson assortativity coefficient are not invariant under network renormalization, so a validated scale-independent diagnostic of hub–hub repulsion remained open even though the model exponent $d_e$ characterizes anticorrelation strength scale-invariantly within the growth model; empirical fractal exponents are hard to determine because the small-world property restricts the range of box sizes, and large exponents (e.g. $d_e=4.5$ for the E. coli metabolic network) may be indistinguishable from exponential (non-fractal) behavior; and the evolutionary claim that robustness selection drives real networks into the fractal regime remained causally undemonstrated. All three cores remain open in the literature: local disassortativity has been shown to be neither necessary nor sufficient for fractality, while long-range correlation formalisms and geometric renormalization provide partial frameworks but no accepted invariant diagnostic; scaling theory and box-covering algorithms advanced, but estimator consistency near the fractal–small-world boundary remains unresolved; and real-network studies explain self-similar growth mechanistically without establishing robustness-selection causation.

Background

Many real networks — the World Wide Web, protein–protein interaction networks, and metabolic networks — are simultaneously scale-free and modular. A scale-free network has a degree distribution (the degree $k$ of a node is its number of links) following a power law $P(k)\sim k^{-\gamma}$; a modular network decomposes into groups of densely connected nodes. Song, Havlin and Makse (Nature 433, 392, 2005) showed that such networks can be self-similar or fractal: covering the network with boxes of diameter $\ell_B$ (the box-covering renormalization, where each box becomes a single node of a renormalized network) yields a number of boxes $N_B\sim \ell_B^{-d_B}$, defining a fractal (box) dimension $d_B$, while scale-free non-fractal networks such as the Internet at the router level are instead small-world — their diameter grows only logarithmically with size, so renormalization quickly collapses them to one box. In 'Origins of fractality in the growth of complex networks' (Nature Physics 2, 275–281, 2006) the same authors proposed a growth mechanism that unifies both cases: at each step node number, hub degree, and diameter grow multiplicatively by factors $n$, $s$, and $a$ (giving $d_B=\ln n/\ln a$, $d_k=\ln s/\ln a$, $\gamma=1+\ln n/\ln s$), and new inter-module links are attached either hub-to-hub (Mode I, probability $e$, producing hub–hub attraction and small-world non-fractal topology) or via non-hubs (Mode II, probability $1-e$, producing hub–hub repulsion and fractal topology). The strength of this 'repulsion' (disassortativity — the tendency of high-degree hubs to connect to low-degree nodes rather than to each other) across all length scales is the proposed hallmark of fractality. The authors supported an evolutionary reading: in model networks matched in size, degree exponent, link number, loops, and clustering, the fractal modular topology survives targeted attacks on the largest hubs far better ($f_c\approx 0.09$ vs $\approx 0.02$ removed hubs to collapse), suggesting that natural selection or design pressure favors fractal architectures. The paper explicitly acknowledged its limitations, which define the open agenda: (i) the Pearson assortativity coefficient $r$ (the standard scalar summary of degree–degree correlation along edges) is not invariant under renormalization, so no scale-independent measure of hub–hub repulsion was validated for empirical networks, even though the model exponent $d_e$ characterizes anticorrelation strength scale-invariantly within the growth model; (ii) empirical exponent determination is severely limited because the small-world property restricts the usable range of $\ell_B$, and large exponents such as $d_e=4.5$ for the E. coli metabolic network (where $E(\ell_B)\equiv n_h(\ell_B)/k_B(\ell_B)$ is the hub-repulsion ratio of links from a box to hubs of neighbouring boxes to the box degree, whose power-law decay $E(\ell_B)\sim\ell_B^{-d_e}$ defines $d_e$, a distinct object from the box-degree exponent $d_k=\ln s/\ln a$) may be empirically indistinguishable from exponential decay, i.e. from the non-fractal limit; and (iii) the robustness-based evolutionary interpretation was suggestive, not causally demonstrated in real evolutionary histories. Later work sharpened rather than closed these gaps: Rozenfeld, Song and Makse (Phys. Rev. Lett. 104, 025701, 2010) formalized the small-world–fractal crossover as renormalization-group flow between fixed points; Fujiki, Mizutaka and Yakubo (Eur. Phys. J. B 90, 126, 2017) showed that maximal disassortativity with a fixed degree sequence mostly produces non-fractal networks, so local anticorrelation is neither necessary nor sufficient for fractality; Fujiki, Takaguchi and Yakubo (Phys. Rev. E 97, 062308, 2018) subsequently gave a general formalism for long-range, shortest-path-resolved degree correlations; García-Pérez, Boguñá and Serrano (Nature Physics 14, 583, 2018) introduced geometric renormalization — renormalization via embeddings in hidden hyperbolic metric spaces (underlying geometric spaces inferred from network topology in which similar nodes are close) — under which real scale-free networks show geometric scaling, a parallel resolution track whose relation to box-covering repulsion diagnostics remains to be consolidated; and Fronczak et al. (Sci. Rep. 14, 9079, 2024) built a consistent scaling theory of fractal networks with seven exponents of which only three are independent, extending but not empirically de-risking exponent measurement.

Problem Statement

Within the renormalization growth theory of Song–Havlin–Makse for modular scale-free networks, establish a validated, scale-consistent account of network fractality, specifically: (1) construct and validate a renormalization-invariant measure of hub–hub correlation — one preserved under box-covering renormalization and capable of separating fractal from non-fractal modular scale-free networks — given that the Pearson assortativity coefficient is not renormalization-invariant, that the source's scale-invariant exponent $d_e$ characterizes repulsion strength within the growth model but its empirical determination from the restricted scaling range is the separate estimation problem addressed in (2), and that local disassortativity has been shown to be neither necessary nor sufficient for fractality; (2) develop statistically reliable inference of the fractal exponents (at minimum $d_B$ and $d_e$) from empirical networks despite the small-world property restricting the accessible range of box sizes $\ell_B$, including a defensible decision rule between large finite exponents (as $d_e=4.5$ for the E. coli metabolic network) and exponential behavior at the non-fractal limit; and (3) test the evolutionary conjecture causally — determine, from evidence on how real networks evolve, whether growth under robustness pressures (e.g. resilience to targeted hub removal) drives networks into the Mode-II-dominated fractal regime of the $(n,s,a,e)$ model, as opposed to fractality arising from geometric or other growth constraints.

References

1. Song, C., Havlin, S., Makse, H. A. (2006). Origins of fractality in the growth of complex networks. Nature Physics 2, 275–281. doi:10.1038/nphys266, https://doi.org/10.1038/nphys266 (author preprint: arXiv:cond-mat/0507216, https://arxiv.org/abs/cond-mat/0507216)
2. Song, C., Havlin, S., Makse, H. A. (2005). Self-similarity of complex networks. Nature 433, 392–395. doi:10.1038/nature03248, https://doi.org/10.1038/nature03248
3. Song, C., Gallos, L. K., Havlin, S., Makse, H. A. (2007). How to calculate the fractal dimension of a complex network: the box covering algorithm. Journal of Statistical Mechanics: Theory and Experiment 2007, P03006. doi:10.1088/1742-5468/2007/03/P03006, https://doi.org/10.1088/1742-5468/2007/03/P03006
4. Rozenfeld, H. D., Song, C., Makse, H. A. (2010). Small-world to fractal transition in complex networks: a renormalization group approach. Physical Review Letters 104, 025701. doi:10.1103/PhysRevLett.104.025701, https://doi.org/10.1103/PhysRevLett.104.025701
5. Fujiki, Y., Mizutaka, S., Yakubo, K. (2017). Fractality and degree correlations in scale-free networks. The European Physical Journal B 90, 126. doi:10.1140/epjb/e2017-80031-x, https://doi.org/10.1140/epjb/e2017-80031-x
6. Fujiki, Y., Takaguchi, T., Yakubo, K. (2018). General formulation of long-range degree correlations in complex networks. Physical Review E 97, 062308. doi:10.1103/PhysRevE.97.062308, https://doi.org/10.1103/PhysRevE.97.062308
7. García-Pérez, G., Boguñá, M., Serrano, M. Á. (2018). Multiscale unfolding of real networks by geometric renormalization. Nature Physics 14, 583–589. doi:10.1038/s41567-018-0072-5, https://doi.org/10.1038/s41567-018-0072-5
8. Zheng, M., García-Pérez, G., Boguñá, M., Serrano, M. Á. (2021). Scaling up real networks by geometric branching growth. Proceedings of the National Academy of Sciences 118, e2018994118. doi:10.1073/pnas.2018994118, https://doi.org/10.1073/pnas.2018994118
9. Schneider, C. M., Moreira, A. A., Andrade, J. S., Havlin, S., Herrmann, H. J. (2011). Mitigation of malicious attacks on networks. Proceedings of the National Academy of Sciences 108, 3838–3841. doi:10.1073/pnas.1009440108, https://doi.org/10.1073/pnas.1009440108
10. Fronczak, A., Fronczak, P., Samsel, M. J., Makulski, K., Łepek, M., Mrowinski, M. J. (2024). Scaling theory of fractal complex networks. Scientific Reports 14, 9079. doi:10.1038/s41598-024-59765-2, https://doi.org/10.1038/s41598-024-59765-2
11. Łepek, M., Makulski, K., Fronczak, A., Fronczak, P. (2025). Beyond traditional box-covering: determining the fractal dimension of complex networks using a fixed number of boxes of flexible diameter. Chaos, Solitons & Fractals 199, 116908. doi:10.1016/j.chaos.2025.116908, https://doi.org/10.1016/j.chaos.2025.116908
12. Bermúdez Gómez, S. (2026). A finite-size scaling theory for fractal dimension estimators in complex networks: inconsistency, decomposition, and adaptive correction. Preprints.org preprint. doi:10.20944/preprints202603.1821.v1, https://doi.org/10.20944/preprints202603.1821.v1
13. Wu, J., Zhu, L., Xu, K., Zheng, M. (2026). Geometric branching growth predicts the evolution of resilience and self-similarity in real networks. Nonlinear Dynamics 114, 663. doi:10.1007/s11071-026-12547-z, https://doi.org/10.1007/s11071-026-12547-z
14. Gabrielli, A., Garlaschelli, D., Patil, S. P., Serrano, M. Á. (2025). Network renormalization. Nature Reviews Physics 7, 203–219. doi:10.1038/s42254-025-00817-5, https://doi.org/10.1038/s42254-025-00817-5