Node-Level Bifractality in Fractal Scale-Free Networks

• Node-level bifractality is a property in fractal scale-free networks where nodes segregate into two groups with distinct local fractal dimensions, typically representing hubs and peripheries.
• It governs the scaling of observables like degree distributions, natural measures, and return probabilities, linking local geometric structures to dynamic behavior.
• The concept clarifies how processes such as diffusion, synchronization, and epidemic spreading differentiate between rapid hub-driven dynamics and slower periphery responses.

Node-level bifractality is a structural property observed in networks that are both scale-free and fractal, particularly fractal scale-free networks (FSFNs). In such systems, the inhomogeneity of local environments is so pronounced that the network cannot be fully characterized by a single fractal dimension; instead, its node set splits into two dominant classes, each defined by a distinct local fractal or spectral dimension. This bifractality at the node level governs the scaling of key structural and dynamical observables—such as the degree, measure, and return probabilities—depending on whether a node is structurally similar to a high-degree hub or to a peripheral, low-degree region. The resulting two-exponent description has ramifications for the analysis of dynamics, diffusion, synchronization, and other processes on FSFNs.

1. Mathematical Framework for Node-Level Bifractality

Node-level bifractality arises when observables associated with nodes—such as natural measure, degree, or local density—scale according to two distinct exponents. In the context of FSFNs, two foundational principles underpin this property:

• Degree Distribution: FSFNs possess a scale-free degree distribution,

$P(k) \propto k^{-\gamma},$

with $\gamma > 2$.

• Box-Counting Fractal Dimension: The box-covering method yields a global fractal dimension $D_f$ through

$N_B(\ell) \sim \ell^{-D_f},$

where $N_B(\ell)$ is the minimal number of subgraphs (“boxes”) of diameter at most $\ell$ needed to cover the network.

To probe local structural inhomogeneities, one introduces the multifractal formalism. For a box $b$ comprising nodes $i \in b$,

$\mu_b(\ell) = \sum_{i\in b} \frac{1}{N}$

defines the box measure. The $q$-th moment “partition sum” is

$\gamma > 2$0

as $\gamma > 2$1. The spectrum $\gamma > 2$2 encodes the scaling of various regions: strict linearity signals monofractality, nonlinearity indicates multifractality, and piecewise linearity with two segments reveals bifractality. At the level of individual nodes, extreme local environments correspond to two exponents: hubs (high degree) and periphery (low degree), formalized as $\gamma > 2$3 and $\gamma > 2$4 respectively (Yamamoto et al., 2023).

2. Emergence of Node-Level Bifractality via Measure–Degree Mapping

In networks derived from multifractal time series or dynamical systems, a direct mapping connects the visit-frequency measure of phase-space partitions with the network node degree. For a trajectory $\gamma > 2$5 on $\gamma > 2$6, let the space be partitioned into boxes of side $\gamma > 2$7. The natural measure of box $\gamma > 2$8 at scale $\gamma > 2$9,

$D_f$0

is empirically found to relate to node degree $D_f$1 via

$D_f$2

with $D_f$3 the box-counting dimension and $D_f$4 an increasing function (typically algebraic or exponential). For a multifractal,

$D_f$5

Reconciling the two,

$D_f$6

implying that as $D_f$7, nodes segregate into bulk ($D_f$8) and hub ($D_f$9) classes, resulting in distinct scaling behaviors for each (Budroni et al., 2016).

3. Analytical Structure: Mass Exponent and Local Fractal Dimensions

The mass exponent $N_B(\ell) \sim \ell^{-D_f},$0, derived from the $N_B(\ell) \sim \ell^{-D_f},$1-moment partition sum, characterizes how different regions of the network scale. Under the bifractality ansatz $N_B(\ell) \sim \ell^{-D_f},$2 (box mass proportional to box degree),

$N_B(\ell) \sim \ell^{-D_f},$3

where $N_B(\ell) \sim \ell^{-D_f},$4 is the global fractal dimension and $N_B(\ell) \sim \ell^{-D_f},$5 is the degree distribution exponent.

From the asymptotic behavior of $N_B(\ell) \sim \ell^{-D_f},$6,

$N_B(\ell) \sim \ell^{-D_f},$7

Here, $N_B(\ell) \sim \ell^{-D_f},$8 characterizes the most “condensed” regions around infinite-degree hubs, while $N_B(\ell) \sim \ell^{-D_f},$9 captures peripheral regions distanced from any hub. This exact two-exponent structure typifies node-level bifractality in FSFNs across hierarchical, stochastic, and critical random-graph models (Yamamoto et al., 2023).

4. Manifestation in Dynamical and Spectral Properties

Node-level bifractality is directly reflected in the network’s dynamical observables. The spectral and walk dimensions, which control diffusion and return probabilities, also separate into two classes:

• Walk Dimension $N_B(\ell)$0: Governs mean displacement scaling,

$N_B(\ell)$1

Empirically, $N_B(\ell)$2 is invariant under choice of starting node.

• Spectral Dimension $N_B(\ell)$3: Sets the decay of return probability,

$N_B(\ell)$4

but $N_B(\ell)$5 bifurcates:

$N_B(\ell)$6

For generator models and critical random graphs, explicit expressions for $N_B(\ell)$7, $N_B(\ell)$8, $N_B(\ell)$9, and $\ell$0 are provided in terms of construction parameters and $\ell$1 (Yakubo et al., 2024).

The table summarizes the bifractal spectral structure for two FSFN classes:

Network Type	$\ell$2	$\ell$3
Generator model (FSFN)	$\ell$4	$\ell$5
Critical scale-free giant comp.	$\ell$6 ($\ell$7)	$\ell$8 ($\ell$9)

Starting from a hub node ($b$0), the return probability decays more slowly compared to a peripheral node ($b$1), indicating fundamentally different return statistics despite the uniform propagation speed set by $b$2.

5. Node-Level Bifractality and Conditions for Scale-Free Behavior

The emergence of node-level bifractality is tightly coupled to the scale-free structure and the multifractal spectrum of the underlying dynamical or empirical network. Specifically, if the measure–degree relation takes the algebraic form $b$3, the degree distribution $b$4 acquires a power-law tail under the criterion

$b$5

where $b$6 is the box-counting dimension and $b$7 (Budroni et al., 2016). Genuine multifractality ($b$8) ensures $b$9 and hence the scale-free regime.

If $i \in b$0 grows exponentially, the degree distribution is exponentially truncated:

$i \in b$1

The precise scaling law for the measure, $i \in b$2, together with the multifractal scaling $i \in b$3, ensures that the observed bifractal splitting at the node level is a direct consequence of the multifractal spectrum of the underlying process.

6. Empirical and Numerical Demonstration

Node-level bifractality is observed both in synthetic network models (deterministic and stochastic hierarchical FSFNs, critical percolation clusters) and in real-world networks (e.g., university web graphs, protein-interaction networks). Numerical computation of the mass exponent $i \in b$4, return probability $i \in b$5, and mean topological displacement $i \in b$6 consistently evidences two distinct scaling regimes corresponding to hub and periphery nodes (Yamamoto et al., 2023, Yakubo et al., 2024). The two-slope structure of $i \in b$7 is reproduced with high precision, and the predicted forms for return probability exponents are validated by direct simulation.

7. Implications for Dynamical Processes and Structural Analysis

Node-level bifractality produces sharp heterogeneity in dynamic processes:

• Random walks probe two effective environments, with shorter return times near hubs and slower “bulk” exploration away from hubs, despite identical displacement scaling.
• Synchronization, diffusion, and epidemic spreading may progress rapidly in hub-centric regions but transition to slower dynamics in peripheral regions.
• Information spreading bifurcates into an initial phase dominated by hub activity ($i \in b$8), followed by slower expansion into low-degree regions ($i \in b$9).

A plausible implication is that bifractality introduces qualitative differences in observables depending on initial conditions (hub vs non-hub), contrasting with homogeneous or monofractal networks. Such insights unify the understanding of network topology and dynamical processes, emphasizing that FSFNs generically exhibit node-level bifractality as an intrinsic organizational principle (Yamamoto et al., 2023, Yakubo et al., 2024).