Mittag-Leffler Graph Dynamics

• Mittag-Leffler graph dynamics are defined as non-Markovian network processes where waiting times follow heavy-tailed distributions, capturing burstiness and memory effects.
• The analytical framework employs fractional calculus, notably Caputo derivatives, to obtain closed-form generating functions and explicit finite-time distributions.
• Applications include modeling anomalous diffusion, dynamic network evolution, and epidemic spreading, demonstrating significant deviations from classical Poissonian models.

Mittag-Leffler graph dynamics refers to a class of non-Markovian network or random-walk processes in which the timing of transitions (such as link activation/deletion or node jumps) follows the heavy-tailed Mittag-Leffler distribution, rather than the standard exponential law associated with Markovian models. This framework facilitates the exact modeling of memory effects, burstiness, and anomalous relaxation phenomena in dynamic networks and random walks on graphs. Analytically tractable formulations employing fractional calculus, particularly via Caputo derivatives, provide explicit descriptions of the finite-time distributions, state evolution, and functional limits of such systems (Georgiou et al., 2015, Michelitsch et al., 2020).

1. Mittag-Leffler Processes on Graphs: Definition and Motivation

Mittag-Leffler graph dynamics generalize classical Markovian dynamical processes on graphs by allowing the inter-event times between transitions to follow a Mittag-Leffler law. For a fixed parameter $\alpha\in (0,1]$ and rate $\lambda>0$, the waiting time density is

$$f(t) = \lambda\, t^{\alpha-1} E_{\alpha,\alpha}\bigl(-\lambda\, t^{\alpha}\bigr)$$

where $E_{\alpha,\beta}(\cdot)$ denotes the two-parameter Mittag-Leffler function. The resulting dynamics are intrinsically non-Markovian, with heavy-tailed statistics and power-law inter-event behavior for $\alpha<1$. This models empirically observed burstiness and memoryful event timings in natural and human-made systems.

In the context of dynamic networks, the random link activation–deletion (RLAD) model is a canonical example: Events at Mittag-Leffler-distributed times randomly flip the state of a uniformly chosen potential link between $N$ nodes (maximum $M = N(N-1)/2$ undirected edges) (Georgiou et al., 2015). More generally, Mittag-Leffler waiting times can subordinate arbitrary discrete-state processes, including random walks on directed or undirected graphs (Michelitsch et al., 2020).

2. Fractional Kolmogorov Forward Equations and Exact Solutions

The non-Markovian nature of Mittag-Leffler dynamics is naturally described by fractional generalizations of Kolmogorov forward (master) equations, utilizing the Caputo fractional derivative of order $\alpha$: $$D_t^\alpha f(t) = \frac1{\Gamma(1-\alpha)}\int_0^t \frac{f'(s)}{(t-s)^\alpha}\,ds$$ For the RLAD network, defining $P_m(t)$ as the probability of $m$ active links at time $t$, the evolution equations read, for $1 \le m \le M-1$,

$$D_t^\alpha P_m(t) = \lambda[(M-m+1)P_{m-1}(t) + (m+1)P_{m+1}(t) - M P_m(t)]$$

with boundary conditions at $m=0$ and $m=M$. For $\alpha=1$, these equations revert to the standard Poissonian (Markov) case.

Closed-form generating function solutions are available, e.g., for RLAD,

$$G(u,t) = [1-q(t) + q(t)u]^M, \qquad q(t) = \frac{1}{2}\left[1 + E_{\alpha,1}(-2\lambda t^\alpha)\right]$$

implying that all links independently undergo non-Markovian telegraph-type (on-off) switching. The time-dependent probability law for the number of active links is thus binomial,

$P_m(t) = \binom{M}{m}\, [q(t)]^{m}[1-q(t)]^{M-m}$

with asymptotic stationarity at $t\to\infty$ yielding $q(\infty)=1/2$ and recovering the equilibrium $B(M,1/2)$ distribution (Georgiou et al., 2015).

3. Mittag-Leffler Subordination and Fractional Laplacian Generators

Mittag-Leffler dynamics on graphs can be constructed by subordination: A base Markov chain or random walk with generator $L$ is time-changed by a renewal process with Mittag-Leffler waiting times, typically corresponding to a fractional Poisson process. On a graph with Laplacian $\Delta$, generalizations via Bernstein functions $f$ (with $L_f = -f(\Delta)$) yield a rich class of admissible transition generators (Michelitsch et al., 2020). For instance, the standard Laplacian on $\mathbb{Z}$ admits functional calculus via Fourier analysis, extending to arbitrary digraphs.

The master equation for the subordinated process takes the form: $$D_t^\alpha \vec{p}(t) = L \vec{p}(t), \quad \vec{p}(0) = \vec{p}_0$$ where $L$ embodies the graph structure and edge weights. For strictly increasing walks with Mittag-Leffler jumps, convolution formulas and explicit series representations involving the Mittag-Leffler and Prabhakar functions obtain.

4. Diffusion Limits and Anomalous Transport

Scaling limits of Mittag-Leffler-driven random walks result in fractional diffusion or transport equations, reflecting anomalous subdiffusive or superdiffusive behavior. In the well-scaled continuous limit ($x = kh$, $h\to 0$, $\lambda = \lambda_0 h^\alpha$ fixed), the state density ${\cal P}(x,t)$ satisfies: $$D_t^\alpha {\cal P}(x,t) = -\xi\,{\cal P}(x,t) + \xi \int_0^x {\cal W}_{ML,\alpha}(x-\tau;\lambda_0) {\cal P}(\tau, t) d\tau$$ where the kernel is the Mittag-Leffler density. The solution admits a convergent Prabhakar-series representation, and connections to general fractional calculus emerge in the diffusion limit (Michelitsch et al., 2020).

This framework unifies a wide class of anomalous transport phenomena on lattices and networks, including heavy-tailed waiting times and long-range jumps.

5. Applications in Networked Dynamics and Epidemic Spreading

Mittag-Leffler graph dynamics furnish an analytically tractable laboratory for studying non-Markovian and bursty network evolution, as well as their effects on processes propagating on such substrates. Simulation studies for the RLAD model with both Mittag-Leffler and Pareto-tailed inter-event times reveal quantitatively excellent agreement in link count statistics, validating the modeling fidelity of the Mittag-Leffler law for general power-law tailed dynamics (Georgiou et al., 2015).

Coupling a concurrent susceptible-infected-susceptible (SIS) epidemic process to the evolving graph state (with fixed-rate infection and recovery but dynamically updating contact structure) exposes the qualitative impact of non-Markovian link dynamics: bursty, heavy-tailed network events slow convergence to equilibrium and substantially increase sustained prevalence compared to the classical Poissonian baseline.

Other applications include:

• Anomalous diffusion on complex networks;
• Fractional stochastic processes in population models;
• Search/navigation algorithms on web or social graphs with heavy-tailed step sizes;
• Modeling damage accumulation or “aging” effects via strictly increasing, counting random walks.

6. Generalizations and Extensions

The renewal-subordination framework underlying Mittag-Leffler graph dynamics extends to arbitrary discrete-state Markov chains whose inter-jump times follow general renewal laws (not necessarily Mittag-Leffler). For any process $X_n$ with transition matrix $Q$, and renewal process $\{S_n\}$ with counting process $N(t)$,

$$\Pr\{X(t)=j\} = \Pr\{T_1>t\}\,\delta_{j,i_0} + \sum_{n=1}^\infty Q^{(n)}_{i_0,j}\, \Pr\{N(t)=n\}$$

This encompasses processes with time-inhomogeneous rates, state- or history-dependent waiting times, or multivariate updates, yielding at minimum a semi-analytic renewal-series solution. However, closed-form Mittag-Leffler–based representations may be lost in these more general cases (Georgiou et al., 2015).

7. Context and Significance for Network Science

The Mittag-Leffler graph dynamics paradigm provides a theoretically robust and computationally efficient toolkit to capture the archetypal properties of bursty and memoryful dynamics on networks, which are pervasive in empirical studies of temporal and adaptive networks. Its analytical solvability via fractional calculus distinguishes it from more heuristic or purely simulation-based heavy-tailed models. A plausible implication is that these techniques can inform both fundamental mathematical theories of anomalous stochastic processes and practical modeling across biology, epidemiology, communications, and critical infrastructure systems. The frameworks developed by Georgiou, Kiss & Scalas (Georgiou et al., 2015) and Michelitsch, Polito & Riascos (Michelitsch et al., 2020) establish Mittag-Leffler graph dynamics as a foundational methodology for non-Markovian network and random walk analysis.