Non-Normal Diffusion & Anomalous Transport

• Non-normal diffusion is a class of transport processes that deviate from classical Brownian motion via nonlinear mean-squared displacement scaling or non-Gaussian propagators.
• It arises from factors like random environments, dynamical heterogeneity, and memory effects, leading to subdiffusive, superdiffusive, or Brownian yet non-Gaussian behaviors.
• Mathematical models including fractional operators, diffusing-diffusivity frameworks, and generalized Fokker–Planck equations enable precise analysis of these anomalous transport phenomena.

Non-normal diffusion refers to a broad set of diffusive behaviors that deviate from the canonical Brownian-Gaussian paradigm in at least one essential aspect: the temporal scaling of the mean-squared displacement (MSD), the statistical structure of the propagator (probability density of displacements), or the spectral content of fluctuations. Traditional normal (Fickian) diffusion is defined by linear-in-time MSD and a Gaussian propagator governed by the heat equation. In contrast, non-normal diffusion encompasses scenarios where the MSD exhibits nonlinear time scaling, the propagator deviates from Gaussianity (while possibly maintaining normal MSD scaling), or both, as a consequence of heterogeneities, environmental fluctuations, disorder, nontrivial memory, or constrained geometries. The landscape of non-normal diffusion includes superdiffusive and subdiffusive regimes, Brownian yet non-Gaussian processes, non-ergodic and multi-scaling phases, and anomalous transients, each with distinct physical origins and mathematical frameworks.

1. Defining Normal and Non-Normal Diffusion

Standard (normal) diffusion in $d$ dimensions is characterized by

• Linear mean-squared displacement: $\langle [x(t)-x(0)]^2 \rangle \propto t$
• Gaussian propagator: $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$
• Governed by the heat equation: $\partial_t P = D \Delta P$

Non-normal diffusion arises in at least two major forms:

• Anomalous diffusion: the MSD scales as $\langle [x(t)]^2 \rangle \propto t^\alpha$, with $\alpha\ne 1$ ($\alpha<1$ for subdiffusion; $\alpha>1$ for superdiffusion). The propagator may be Gaussian (e.g., fractional Brownian motion) or non-Gaussian (e.g., Lévy flights) (Deng et al., 2018, Shin et al., 2014).
• Non-Gaussian normal diffusion: the MSD retains its normal (linear) time scaling, $\langle [x(t)]^2 \rangle \propto t$, but the propagator $P(x,t)$ is non-Gaussian, exhibiting exponential (Laplace) tails or more complex features (cusps, stretched exponential, multi-peaked structure) (Yin et al., 2021, Abe, 2020, Wang et al., 2015, Li et al., 2019).

Distinguishing between these forms requires consideration of both scaling laws and the full statistics of particle displacements.

2. Physical Mechanisms and Microscopic Scenarios

Non-normal diffusion can arise from a variety of physical origins:

2.1 Random Environments and Disorder

• In disordered potentials, the waiting times or jump lengths can acquire broad or divergent distributions, leading to subdiffusive or superdiffusive regimes depending on the statistical tails of the disorder (Salgado-Garcia, 2015, Goychuk et al., 2014).
• Spatial correlations in the disorder induce transient subdiffusion on mesoscales, with a crossover to normal diffusion at macroscales controlled by an ergodicity length (Goychuk et al., 2014).
• Energy landscapes with exponential density of states and spatial correlations display a two-stage transition: a regime of anomalous (superdiffusive) transport precedes the full breakdown to non-equilibrium dispersive motion as temperature is reduced (Novikov, 2017).

2.2 Environmental and Dynamical Heterogeneity

• In complex crowded or fluctuating environments, time-dependent or spatially heterogeneous diffusivities generate Brownian yet non-Gaussian diffusion. The propagator becomes a superposition of Gaussians with varying widths, yielding exponential or even more complex tails (Wang et al., 2015, Santos et al., 2021, Li et al., 2019, Abe, 2020).
• In soft-matter and biological systems, structural rearrangements, compartmentalization, or intermittent trapping (e.g., fluctuating channel geometries) are prominent causes of non-Gaussian displacement statistics with normal MSD scaling (Li et al., 2019, Yin et al., 2021, Nayak et al., 19 Feb 2025).

2.3 Mode Coupling and Symmetry-Induced Anomalies

• In one-dimensional interacting systems, such as integrable spin chains or nonlinear lattices, conservation laws and the presence or absence of zero-frequency phonon modes can produce a coexistence of normal and anomalous diffusion at different spatial or temporal scales (Steinigeweg et al., 2012).
• In 1D lattices with preserved or emergent symmetries (e.g., parity momentum), superdiffusive energy transport can occur even when momentum is not globally conserved, with the anomalous exponent directly linked to the nontrivial conserved quantity and the phonon spectrum (Yan et al., 2019).

2.4 Nonlinear and Nonlocal Effects

• Nonlinear dependencies of mobility or diffusivity on concentration or other fields generate transient anomalous diffusion (e.g., subdiffusive scaling $\langle [x(t)-x(0)]^2 \rangle \propto t$0 crossing over to normal as the system homogenizes) purely via local interactions, in contrast to nonlocal (fractional) models which generate permanent anomalies (Barreiro et al., 12 Jan 2026).

2.5 Heterogeneous Ensembles and Superstatistical Mixtures

• Populations of random walkers with distributed, possibly time-dependent, diffusivities (superstatistics) provide a universal mechanism for non-Gaussian normal diffusion, with the propagator determined by the Laplace transform of the underlying diffusivity distribution (Santos et al., 2021, Abe, 2020, Baldovin et al., 2019).

3. Mathematical Formulations and Theoretical Frameworks

Key mathematical approaches underlying non-normal diffusion include:

3.1 Diffusing-Diffusivity and Hierarchical Fokker-Planck Models

• The joint Fokker-Planck equation for $\langle [x(t)-x(0)]^2 \rangle \propto t$1, with fast $\langle [x(t)-x(0)]^2 \rangle \propto t$2 and slow $\langle [x(t)-x(0)]^2 \rangle \propto t$3 variables, generates marginal propagators that are mixtures of Gaussians with arbitrary $\langle [x(t)-x(0)]^2 \rangle \propto t$4 (Abe, 2020).
• Under exponential $\langle [x(t)-x(0)]^2 \rangle \propto t$5, the propagator becomes a Laplace distribution in $\langle [x(t)-x(0)]^2 \rangle \propto t$6, producing the signature "Brownian yet non-Gaussian" statistics (Abe, 2020, Wang et al., 2015).

3.2 Fractional and Nonlocal Operators

• Fractional Laplacians $\langle [x(t)-x(0)]^2 \rangle \propto t$7, anisotropic or tempered to control divergence, serve as generators for generalized, non-normal transport processes, encompassing superdiffusive Lévy flights and their finite-variance relatives (Deng et al., 2018).
• Anisotropy and tempering allow modeling of transport with directional bias, truncation of jumps, and heterogeneous media.

3.3 Generalized Langevin, Memory Functions, and Einstein Relations

• Anomalous diffusion in fluids and solids can be modeled by generalized Langevin equations with memory kernels; a generalized Einstein relation links time-dependent friction and diffusion coefficients for arbitrary scaling exponent $\langle [x(t)-x(0)]^2 \rangle \propto t$8 (Shin et al., 2014).

3.4 Superstatistical and Subordinated Processes

• Populations of fractional Brownian walkers with distributed mobility and time-dependent scaling protocols yield self-similar propagators, rescalable to universal functions via Laplace (or Mellin) transforms (Santos et al., 2021).
• Kurtosis, higher cumulants, and the full multi-scaling spectrum $\langle [x(t)-x(0)]^2 \rangle \propto t$9 are directly accessible from the subordination framework (Rebenshtok et al., 2013, Baldovin et al., 2019).

3.5 Generalized Diffusion Equations and Point Transformations

• Point transformations $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$0 recast normal diffusion in $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$1 as anomalous (or multimodal) in $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$2, leading to generalized diffusion equations and nontrivial stationary distributions, including bimodal attractors and multi-peaked profiles (Kouri et al., 2017).

4. Empirical and Simulation Observations

• Experimental and simulation studies confirm robust signatures of non-normal diffusion in diverse systems: colloidal particles in disordered media (Wang et al., 2015, Yin et al., 2021), soft condensed matter under MIPS (Nayak et al., 19 Feb 2025), tracer diffusion in fluctuating channels (Li et al., 2019), and active or anisotropic particles (Yin et al., 2021).
• Simulations show that disorder, non-equilibrium steady states, and slow environmental fluctuations can maintain non-Gaussianity for time scales much longer than the microscopic relaxation, with crossover timescales $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$3 dictated by diffusivity correlation times or mixing dynamics (Baldovin et al., 2019, Yin et al., 2021).
• Energy superdiffusion in specific 1D momentum nonconserving lattice models has been quantitatively linked to emergent conserved quantities associated with phonon modes at Brillouin zone boundaries, as opposed to traditional momentum conservation (Yan et al., 2019).

5. Applications and Design in Complex, Artificial, and Generative Systems

• Recent developments in diffusion-based generative modeling highlight the flexibility obtained by relaxing Gaussian increments: non-normal diffusion models with arbitrary step distributions (e.g., Laplace, uniform) rigorously preserve sampling dynamics (as step size $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$4) while permitting principled control over loss functions, robustness, and expressiveness in learning (Li, 2024).
• In population biology, finance, and information spreading, non-Gaussian kernels, anomalous scaling, and subordination-coupled frameworks enable accurate capture of empirically observed fat-tailed, non-Fickian propagators and their statistics (Santos et al., 2021).

Table: Main Types of Non-Normal Diffusion

Type	MSD Scaling	Propagator Character
Anomalous Diffusion	$P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$5, $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$6	Gaussian or non-Gaussian
Brownian yet Non-Gaussian	$P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$7	Non-Gaussian (e.g., Laplace, cusp)
Multiscaling / Strong Anomaly	Nonlinear $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$8 spectrum	Multifractal, non-normalizable ICD
Non-Gaussian Diffusion Models	Arbitrary (controlled by step law)	Arbitrary (learned, not fixed)

6. Crossover Regimes, Ergodicity, and Unified Perspectives

• Non-normal diffusion is often transient: environmental or dynamic heterogeneity yields non-Gaussian propagators for intermediate times, but the central limit theorem eventually restores Gaussianity for $P(x,t) = (4\pi D t)^{-d/2} \exp( -|x|^2/(4 D t))$9, except in systems with infinite memory or non-ergodicity (Wang et al., 2015, Yin et al., 2021, Li et al., 2019).
• In systems with strong disorder or global constraints (e.g., conservation laws in 1D chains), anomalous scaling and non-Gaussianity can persist indefinitely or be restored by lifting the underlying symmetry (e.g., by adding on-site potentials) (Steinigeweg et al., 2012, Yan et al., 2019).
• The framework of infinite covariant densities (ICDs) provides a rigorous statistical description of the scaling limits and fluctuation statistics in strongly anomalous open systems, generalizing the central limit theorem for the central bulk and non-normalizable tails (Rebenshtok et al., 2013).

7. Mathematical and Practical Implications

• Non-normal diffusion necessitates careful identification of both time-scaling exponents and the full propagator form. Empirical MSD linearity is not sufficient to guarantee Gaussianity; kurtosis, higher moments, or full distribution fitting must be used (Yin et al., 2021, Baldovin et al., 2019).
• The mathematical backbone involves generalized Fokker-Planck equations, fractional/nonlocal operators, spatio-temporal subordination, and spectral methods in non-Cartesian spaces (Deng et al., 2018, Kouri et al., 2017).
• Model design in computational and statistical mechanics (e.g., in generative modeling) now includes arbitrary control over noise increments and corresponding objective functions, demonstrating the decoupling of training loss geometry and asymptotic sampling fidelity (Li, 2024).

---

In summary, non-normal diffusion encompasses a spectrum of phenomena unified by their deviation from the Gaussian/Fickian ideal in either scaling, propagator, or both. Its theoretical description spans random walks with heterogeneous or evolving mobility, nonlocal PDEs, subordination statistics, and symmetry-induced transport anomalies. Its empirical signatures are increasingly observed across physics, chemistry, biology, and artificial systems, with both scientific and technological implications for the understanding, modeling, and control of transport processes in complex environments.