Brownian yet Non-Gaussian Diffusion (BYNGD)

• BYNGD is a transport phenomenon where linear mean-squared displacement coexists with non-Gaussian displacement distributions, highlighting dynamic heterogeneity.
• Diagnostic measures such as kurtosis and tail length scaling distinguish Gaussian cores from exponential or stretched tails, unveiling underlying statistical structures.
• Mechanistic frameworks including superstatistics, random diffusivity models, and correlated landscapes explain BYNGD across diverse systems like soft matter, biological media, and complex fluids.

Brownian yet non-Gaussian diffusion (BYNGD) denotes stochastic transport in which the mean-squared displacement grows linearly in time, as in ordinary Brownian motion, while the displacement statistics are non-Gaussian over experimentally relevant time scales. In its canonical form, BYNGD combines Fickian scaling,

$\langle x^2(t)\rangle \propto t,$

with a displacement probability density that deviates from

$$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$

A substantial literature now treats BYNGD as a broad class rather than a single mechanism: in some systems the position PDF itself is non-Gaussian, often with exponential or Laplace-like tails, whereas in others the long-time position PDF is nearly Gaussian and the non-Gaussianity is concentrated in increment statistics (Chechkin et al., 2016, Białas et al., 2020, Metzler et al., 2022).

1. Definition and scope

BYNGD is distinguished from both standard Brownian diffusion and anomalous diffusion by the combination of two criteria. First, the MSD is Brownian: $\langle x^2(t)\rangle \propto t.$ Second, the displacement PDF, self van Hove function, or increment PDF is non-Gaussian. This contrasts with anomalous diffusion, for which

$$\langle x^2(t)\rangle \propto t^\beta,\qquad \beta\neq 1,$$

and with standard Brownian motion, for which both the MSD and the displacement PDF are Gaussian in the usual sense (Baldovin et al., 2019, Metzler et al., 2022).

The term covers several statistically distinct situations. In the most common usage, the displacement PDF has a Gaussian core and exponential or stretched tails. However, there are also systems in which the long-time position PDF is “very close to Gaussian,” while the increment PDF remains non-Gaussian with exponentially decaying tails; this variant has been established for overdamped motion in periodic potentials driven by nonequilibrium noise (Białas et al., 2020). A further refinement is that non-Gaussianity need not be encoded only in tails: in driven periodic systems the excess kurtosis can pass through platykurtic, leptokurtic, and mesokurtic stages while the asymptotic process is still described as Brownian yet non-Gaussian because the distribution is only Gaussian-like (Marchenko et al., 23 Jan 2025).

Empirically, BYNGD has been reported across biological, soft-matter, glassy, active, and driven systems, including colloids in complex media, lipid tubes and membranes, gels, active matter, living cells, Lennard–Jones glass-formers, periodic potentials, gas mixtures with strong mass contrast, and diffusion near surfaces (Metzler et al., 2022, Miotto et al., 2019, Nakai et al., 2023).

2. Statistical structure and diagnostics

A central mathematical representation of BYNGD is the mixture of Gaussian propagators over a distribution of diffusivities: $$P(x,t)=\int_0^\infty f(D)\,\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right)\,dD.$$ In superstatistical language, each trajectory is locally Brownian with a quasi-static diffusivity, while the ensemble is heterogeneous. In diffusing-diffusivity models, the same structure emerges dynamically because the diffusivity itself is a stochastic process (Chechkin et al., 2016, Sposini et al., 2018).

The simplest diagnostic remains the coexistence of linear MSD with a non-Gaussian displacement PDF. Higher-order diagnostics are required because MSD alone cannot discriminate Gaussian from non-Gaussian transport. A standard measure is the kurtosis or excess kurtosis,

$$K(t)=\frac{\langle [x(t)-\langle x(t)\rangle]^4\rangle}{\langle [x(t)-\langle x(t)\rangle]^2\rangle^2}-3,$$

or equivalent non-Gaussian parameters based on fourth moments (Białas et al., 2020, Metzler et al., 2022). In random-diffusivity Langevin models, ensemble-averaged time-averaged MSDs remain normal even when ensemble MSD can be anomalous, and the scatter of the dimensionless amplitude is controlled by the time average of $D(t)$; this makes TAMSD-based ergodicity diagnostics informative but not sufficient on their own (Wang et al., 2020).

A second diagnostic is the tail length scale of the displacement PDF. In glass-forming systems, the radial distribution $P(r;t)$ has a Gaussian core and an exponential tail,

$$P(r;t)\sim \exp\!\left(-\frac{r}{\lambda(t)}\right),$$

and the scaling of $\lambda(t)$ can itself carry mechanism-specific information. In Lennard–Jones systems, $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$0 rather than universally $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$1, while tetrahedral gelling systems show a temperature-dependent exponent (Miotto et al., 2019).

A third diagnostic is whether the non-Gaussianity resides in positions or increments. Periodic-potential systems driven by Poisson shot noise can have a Gaussian-like long-time position PDF with non-Gaussian increments, whereas near-wall diffusion shows non-zero fourth cumulant and Brownian MSD but Gaussian rather than exponential tails in the displacement distribution (Białas et al., 2020, Alexandre et al., 2022). This already excludes any universal identification of BYNGD with Laplace tails.

3. Mechanistic frameworks

The current literature supports several non-equivalent mechanisms for BYNGD.

Mechanism	Characteristic feature	Representative paper
Superstatistics / diffusing diffusivity	Gaussian propagators mixed over static or slowly varying diffusivities; crossover to Gaussian at long times	(Chechkin et al., 2016)
Random-coefficient / random-diffusivity Langevin models	Linear MSD with non-Gaussian PDFs from random variance structure; rcAR equivalence in discrete time	(Ślęzak et al., 2019)
Slowly varying heterogeneous landscapes	BYNGD depends on sampling protocol; equilibrated Ito dynamics is the paradigmatic case	(Postnikov et al., 2018)
Correlated diffusivity landscapes	Persistent central peak from strong spatiotemporal correlations and high-order temporal correlations	(Pacheco-Pozo et al., 2023)
Polymerization-induced diffusivity fluctuations	$$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$2 with fluctuating chain length $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$3; center-of-mass BYNGD	(Baldovin et al., 2019)
Nonequilibrium shot noise in periodic potentials	Brownian motion with non-Gaussian increments and colossal diffusion enhancement	(Białas et al., 2021)
Heavy-light gas mixtures	Ensemble averaging over quasi-constant particle speeds in a Lorentz-gas regime	(Nakai et al., 2023)
Long-range molecular interactions	Laplacian displacements from interaction-induced forces and phase-transition-assisted crossover	(Chacón et al., 2022)
Near-surface diffusion	Height-dependent $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$4 generates Brownian but non-Gaussian parallel transport with Gaussian tails	(Alexandre et al., 2022)

In mesoscopic formulations, BYNGD is often interpreted through either superstatistics or subordination. The minimal diffusing-diffusivity theory writes the particle coordinate as Brownian motion in a random operational time set by the integrated diffusivity, which establishes a direct bridge between dynamic diffusivity fluctuations and superstatistical Gaussian mixtures (Chechkin et al., 2016). A closely related statistical viewpoint maps Langevin equations with random damping or diffusion coefficients onto random-coefficient autoregressive processes, making available codifference, PACF, hidden-Markov, and related time-series tools for detecting BYNGD in trajectory data (Ślęzak et al., 2019).

Microscopic realizations demonstrate that fluctuating diffusivity need not be postulated phenomenologically. In polymerization-induced BYNGD, the center-of-mass diffusivity $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$5 fluctuates because polymer length undergoes a birth-death process under reversible monomer addition and removal, furnishing a microscopic realization of diffusing diffusivity and superstatistics (Baldovin et al., 2019). In heavy-light gas mixtures, each trajectory behaves as a Lorentz gas with quasi-constant speed on intermediate times, but the Maxwell–Boltzmann speed ensemble produces a distribution of effective diffusivities and hence BYNGD (Nakai et al., 2023). In nonequilibrium periodic potentials, by contrast, non-Gaussianity can be induced by impulsive external forcing rather than by spatial or temporal diffusivity fluctuations (Białas et al., 2020, Białas et al., 2021).

4. Temporal structure, crossover, and scaling laws

A recurring theme is the separation between short-time non-Gaussianity and long-time Gaussianization. In the minimal diffusing-diffusivity model, superstatistics is exact at times shorter than the diffusivity correlation time, while at longer times the displacement PDF crosses over to a Gaussian with effective diffusivity equal to the stationary mean of $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$6; the MSD is linear throughout (Chechkin et al., 2016). This short-time/long-time dichotomy is the template against which many later mechanisms are compared.

Microscopic models supply explicit crossover scales. In polymerization-induced BYNGD, the autocorrelation decay time of the chain length,

$$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$7

sets the persistence time of a given diffusivity and therefore the crossover from Brownian yet non-Gaussian dynamics to ordinary Brownian motion (Baldovin et al., 2019). In driven periodic systems, the excess kurtosis can evolve non-monotonically, being first negative, then positive, and then tending to zero. The same parameter regime displays transient superdiffusion followed by subdiffusion before the onset of Brownian diffusion, establishing a correlation between transient anomalous diffusion and non-Gaussianity during the approach to nonequilibrium (Marchenko et al., 23 Jan 2025).

Length-scale evolution refines this picture. In isotropic Lennard–Jones systems, the characteristic exponential-tail scale obeys

$$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$8

with explicit forms $$P(x,t)=\frac{1}{\sqrt{4\pi Dt}}\exp\!\left(-\frac{x^2}{4Dt}\right).$$9 in 2D, $\langle x^2(t)\rangle \propto t.$0 in 3D, and $\langle x^2(t)\rangle \propto t.$1 in 4D, while tetrahedral gelling systems display a temperature-dependent exponent that can approach $\langle x^2(t)\rangle \propto t.$2 in the deep Arrhenius regime (Miotto et al., 2019). This directly contradicts the naive expectation that $\langle x^2(t)\rangle \propto t.$3 should always scale like the MSD. The implication is that BYNGD cannot be characterized solely by the existence of an exponential tail; the growth law of that tail is itself mechanism-sensitive.

Sampling protocol also matters. In diffusion on slowly varying random landscapes, BYNGD is described as extremely unlikely under homogeneous sampling from system preparation onward, but it can arise under equilibrated conditions in the Ito interpretation, where the process crosses from Laplace-like to Gaussian displacements while preserving a constant Brownian MSD slope (Postnikov et al., 2018). In correlated diffusivity landscapes, the convergence route to Gaussian is highly unusual: under appropriate rescaling, the central peak narrows and stays sharp rather than simply smoothing and lowering, and this effect disappears when the underlying spatiotemporal correlations are destroyed (Pacheco-Pozo et al., 2023).

5. Experimental systems and applications

The experimental identification of BYNGD typically requires more than the MSD. The standard protocol is to combine particle tracking with the self van Hove function or increment PDF, and then quantify non-Gaussianity through kurtosis, the non-Gaussian parameter, or related fourth-order cumulants. In polymerizing systems, the proposed observables are the center-of-mass MSD, the van Hove function, and the kurtosis, together with comparison to the predicted crossover time $\langle x^2(t)\rangle \propto t.$4 (Baldovin et al., 2019).

Near-wall diffusion provides a particularly clean realization. A colloid diffusing near a wall or in a planar channel has height-dependent local diffusivity parallel to the wall. The parallel displacement remains Brownian in the sense of its variance, but has a non-zero fourth cumulant. Theory, numerics, and experiments agree quantitatively, and the tail analysis shows Gaussian rather than exponential tails, which imposes a strong constraint on generic BYNGD narratives (Alexandre et al., 2022).

Driven systems reveal functional consequences of non-Gaussianity. In periodic potentials with nonequilibrium Poisson shot noise, the non-Gaussian increment statistics are tied to colossal diffusion enhancement, surpassing the giant-diffusion regime of deterministic tilting and strongly affecting first-arrival problems and diffusion-limited reactions (Białas et al., 2020). In active Brownian ratchets driven by exponentially correlated Poisson noise, sparse finite-time bursts produce Brownian yet non-Gaussian displacement statistics and are more efficient for transport and diffusion enhancement than active Ornstein–Uhlenbeck noise (Paneru et al., 2021). These results connect BYNGD to active fluctuations in living cells and to transport optimization in driven mesoscopic systems.

Gas mixtures and active or crowded media broaden the application range. In heavy-light gases, KMC results and Lorentz-gas-based theory show a Brownian MSD and non-Gaussian displacement distribution due to speed heterogeneity, with a non-zero plateau of the non-Gaussian parameter over intermediate times (Nakai et al., 2023). In glasses and gels, the tail scale $\langle x^2(t)\rangle \propto t.$5 diagnoses rare hopping and dynamical heterogeneity (Miotto et al., 2019). In active, cellular, and soft-matter systems more generally, the recurring experimental motif is the same: a Brownian-looking MSD masks statistically heterogeneous transport modes that remain visible in the full displacement or increment distribution (Metzler et al., 2022).

6. Conceptual distinctions, misconceptions, and open problems

A first misconception is that BYNGD always means an exponential displacement PDF. Exponential or Laplace-like tails are common, but not universal. Near surfaces the tails are Gaussian (Alexandre et al., 2022); with fat-tailed shot-noise amplitudes they can be power-law (Białas et al., 2021); in heavy-light gases the asymptotic tail derived from speed averaging is described as neither purely exponential nor stretched Gaussian (Nakai et al., 2023). The generic content of BYNGD is therefore the mismatch between second-moment Brownianity and full-distribution Gaussianity, not a specific tail law.

A second misconception is that non-Gaussianity must reside in the position PDF. Several nonequilibrium periodic-potential models exhibit long-time Gaussian or Gaussian-like position PDFs while the increment PDFs remain distinctly non-Gaussian and retain exponential tails (Białas et al., 2020, Białas et al., 2021). This variant sharpens the distinction between ensemble spreading and finite-lag transport events.

A third misconception is that vanishing excess kurtosis guarantees Gaussian statistics. In the periodically driven nonequilibrium model, the excess kurtosis tends to zero while the position PDF remains only Gaussian-like because the periodic potential imprints a fine comb structure on the distribution (Marchenko et al., 23 Jan 2025). A similar lesson follows from correlated diffusivity landscapes, where the wings can be captured by correlated CTRW approximations but the sharp central peak requires higher-order temporal correlations that are invisible to low-order moment diagnostics (Pacheco-Pozo et al., 2023).

Open problems remain substantial. Quenched heterogeneity versus annealed diffusivity fluctuations remains a central distinction, especially because equilibrated Ito diffusion in slowly varying landscapes is singled out as the paradigmatic heterogeneous-medium realization of BYNGD, whereas homogeneous sampling is said to make BYNGD extremely unlikely (Postnikov et al., 2018). The dimension- and temperature-dependent scaling of the non-Gaussian length scale in glass-formers still constrains existing theories (Miotto et al., 2019). More broadly, the field continues to ask which observables are minimally sufficient to infer mechanism: full displacement PDFs, increment PDFs, TAMSD scatter, codifference, and higher-order temporal correlations all appear necessary in different settings (Ślęzak et al., 2019, Wang et al., 2020).

BYNGD is therefore best understood not as a single anomaly but as a family of transport regimes in which normal diffusion at the level of the MSD coexists with hidden dynamical heterogeneity, fluctuating diffusivity, intermittent forcing, or rare-event structure that remains legible in full distributional statistics.