Run-and-Tumble Dynamics in Active Matter

• Run-and-tumble dynamics is a stochastic process where persistent runs alternate with tumble events, modeling bacterial motility and active matter transport.
• It employs telegraph processes and probabilistic reorientation rules to capture ballistic behavior at short times and diffusive transport at long times.
• Extensions addressing inertial effects, confinement, and engineered realizations demonstrate its role in optimizing chemotaxis and designing novel active systems.

Run-and-tumble dynamics denotes a class of active stochastic processes in which persistent runs alternate with reorientation events that reset, partially reset, or otherwise reorganize the direction of motion. In the canonical bacterial setting, a swimmer runs at speed $v$, tumbles with rate $\lambda$, and may spend a finite time $\tau$ in a tumble state before resuming motion; the resulting trajectory is ballistic at short times and diffusive at long times in the dilute limit (Paoluzzi et al., 2013). Across recent work, the same structure appears in confined single-particle models, interacting suspensions, inertial and multistate generalizations, thermodynamic reconstructions, and robotic realizations, so “run-and-tumble” now refers both to a specific bacterial motility mechanism and to a broader nonequilibrium transport paradigm (Elgeti et al., 2015, Shaebani et al., 2021).

1. Canonical kinematics and stochastic formulations

The minimal run-and-tumble particle alternates between a running state and a reorientation state. In one common formulation, the active drive is a telegraph process, as in

$$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$

with $\sigma(t)\in\{+1,-1\}$ and

$$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$

so that $\gamma^{-1}$ is the persistence time (Frim et al., 2023). In two-dimensional bacterial models, the corresponding physical parameters are the self-propulsion speed $v$, the tumbling rate $\lambda$, and the tumble duration $\tau$, with mean run time $\lambda$0 (Paoluzzi et al., 2013).

A second standard parametrization emphasizes run length rather than run time. In confinement, a run has duration $\lambda$1 and length

$\lambda$2

and, in the low-density model between parallel walls, the behavior depends on $\lambda$3 and $\lambda$4 only through $\lambda$5 or its mean for distributed run lengths (Elgeti et al., 2015). The wall-normal coordinate then updates as

$\lambda$6

with $\lambda$7 redrawn uniformly on the circle in 2D or the sphere in 3D after each tumble (Elgeti et al., 2015).

Directional persistence after tumbles is not universal. In discrete-time two-state models, the run state is characterized by a persistence parameter

$\lambda$8

which reduces to $\lambda$9 for left-right symmetric turning-angle distributions; $\tau$0 corresponds to no directional memory and $\tau$1 to perfectly straight motion (Shaebani et al., 2021). In chemotaxis models, persistence is similarly encoded through the post-tumble angular distribution, and the optimization problem is naturally posed in the two-dimensional parameter space of base tumble frequency $\tau$2 and persistence parameter $\tau$3 (Kirkegaard et al., 2017).

These formulations already show that run-and-tumble dynamics is not a single model but a family of piecewise-ballistic stochastic processes distinguished by how runs end, how tumbles reorient, and whether the tumble state is instantaneous, finite, resting, reversing, or internally driven.

2. Transport diagnostics and asymptotic regimes

For a free run-and-tumble particle with finite tumble duration, the mean-square displacement takes the form

$\tau$4

with diffusivity

$\tau$5

and ballistic-to-diffusive crossover time

$\tau$6

(Paoluzzi et al., 2013). This is the standard transport signature: persistent motion at short times, normal diffusion at long times.

More general two-state models produce multiple transient regimes before this asymptotic limit. In the discrete-time run-and-tumble framework with a fast persistent run state and a slow isotropic tumble state, the approach to diffusion is governed by two independent relaxation times,

$\tau$7

associated respectively with relaxation of state occupancy and decay of directional persistence (Shaebani et al., 2021). This separation allows superdiffusive, diffusive-looking, and effectively subdiffusive windows to coexist before the final diffusive regime is reached.

A central methodological point is that linear growth of the mean-square displacement is not sufficient to identify diffusion. The displacement distribution must also become Gaussian. In the analysis of dilute bacterial models, the relevant object is the van Hove function $\tau$8, and a stringent diagnostic is the excess kurtosis

$\tau$9

which vanishes only for a Gaussian radial distribution in two dimensions (Villa-Torrealba et al., 2020). Models with partial reorientation, run-and-reverse dynamics, or internal biochemical control can exhibit $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$0 while $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$1 remains large for long times, so the transport is only asymptotically diffusive in the full distributional sense (Villa-Torrealba et al., 2020).

At the level of density fluctuations rather than single-particle displacement, interacting and noninteracting run-and-tumble systems are often characterized by the intermediate scattering function $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$2 and the dynamic structure factor $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$3, which resolve how density modes at length scale $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$4 relax in time or frequency (Paoluzzi et al., 2013). These observables are especially useful once many-body interactions make direct trajectory interpretation ambiguous.

3. Confinement, walls, and trapping

Confinement converts run-and-tumble persistence into structured nonequilibrium steady states. For a particle between hard walls at $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$5, the fundamental quantity is the tumbling density $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$6, the probability density for a tumble to occur at wall-normal position $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$7 (Elgeti et al., 2015). With constant run length $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$8, the transfer kernel has a hard cutoff at $$\dot{x}=-\mu \partial_x U + v_0 \sigma(t) + \eta(t),$$9, which generates wall $\sigma(t)\in\{+1,-1\}$0-accumulation, depletion layers near the wall, and peaks at distances of order $\sigma(t)\in\{+1,-1\}$1 from the wall. With exponentially distributed run lengths,

$\sigma(t)\in\{+1,-1\}$2

those sharp structures disappear because the transfer kernel becomes continuous (Elgeti et al., 2015). In wide channels, the surface accumulation obeys the universal law

$\sigma(t)\in\{+1,-1\}$3

with $\sigma(t)\in\{+1,-1\}$4 an order-one accumulation factor (Elgeti et al., 2015). The same work explicitly notes that RTPs and ABPs are not equivalent near walls, even when they can look similar in bulk or weak fields (Elgeti et al., 2015).

In a circular domain, geometric invariance can survive remarkably broad modifications of the run statistics. For a disk of radius $\sigma(t)\in\{+1,-1\}$5, isotropic tumble angles together with elastic reflections recover the Mean Path Length Theorem,

$\sigma(t)\in\{+1,-1\}$6

and this remains valid for Gaussian, uniform, fixed, and exponential step-length distributions under the conditions studied (Zamora et al., 2024). The theorem fails, however, under non-uniform angular distributions, random boundary scattering, or memory processes, because these destroy homogeneous isotropic wall sampling (Zamora et al., 2024).

A different confined setting introduces trapping regions where active particles irreversibly enter a blocked state at rate $\sigma(t)\in\{+1,-1\}$7. When all space is trapping, the survival probability is exactly exponential,

$\sigma(t)\in\{+1,-1\}$8

so the mean trapping time is simply the inverse trapping rate (Angelani, 2023). In a semi-infinite geometry, by contrast,

$\sigma(t)\in\{+1,-1\}$9

while for a finite trapping interval the trapping-time density develops a $$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$0 tail and the mean trapping time is undefined (Angelani, 2023).

Near real surfaces, confinement couples to geometry and hydrodynamics. A single bacterium above a no-slip wall, modeled as two spheres connected by a rod, has surface residence time

$$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$1

and simulations give the geometric scaling

$$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$2

together with an optimal tumble bias $$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$3 for the typical bacterium considered (Leishangthem et al., 29 Jan 2025). Under strong confinement, synthetic thermophoretic Janus ellipsoids can even develop run-and-tumble-like motion without biochemical switching: hydrodynamic wall interactions generate an effective double-well potential for the out-of-plane angle, and noise-driven transitions between the wells define the run time $$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$4 (Anchutkin et al., 2024). A plausible implication is that confinement does not merely perturb run-and-tumble dynamics; in some systems it creates it.

4. Interactions, crowding, and collective renormalization

In interacting active suspensions, the dilute single-particle picture survives only in renormalized form. For a two-dimensional bath of sterically repulsive run-and-tumble swimmers, density fluctuations are analyzed through the dynamic structure factor and fitted with the exact noninteracting theory, but with scale- and density-dependent effective parameters

$$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$5

(Paoluzzi et al., 2013). The robust trends are

$$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$6

with $$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$7 as density increases (Paoluzzi et al., 2013). The mapping works well at moderate densities and intermediate or large $$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$8, but breaks down at high density and especially at small $$\langle \sigma(t')\sigma(t)\rangle = e^{-\gamma |t-t'|},$$9, where genuine many-body correlations cannot be collapsed onto a free-particle process (Paoluzzi et al., 2013). The same paper notes a textual inconsistency in its discussion of the $\gamma^{-1}$0-dependence of $\gamma^{-1}$1; what remains unambiguous is its strong dependence on density (Paoluzzi et al., 2013).

Tagged-particle observables reveal a complementary aspect of crowding. In an active Rouse-like polymer of run-and-tumble monomers, the tagged monomer displays ballistic, superdiffusive $\gamma^{-1}$2, subdiffusive $\gamma^{-1}$3, and diffusive regimes, while non-Gaussianity is confined to times below the persistence time $\gamma^{-1}$4 (Put et al., 2019). In a two-dimensional persistent exclusion lattice gas, by contrast, the tagged particle is superdiffusive at low density, can become subdiffusive at intermediate density, and retains strong non-Gaussian displacement statistics over all simulated times; at intermediate density the tails are approximately Laplace-distributed for sufficiently large times (Put et al., 2019). The proposed mechanism is a dynamically changing environment, with alternating free and blocked episodes that continuously reshape the tagged particle’s effective diffusivity (Put et al., 2019).

Extreme confinement alters the asymptotics even more drastically. In a dense single-file system, a one-dimensional run-and-tumble tracer has exact full statistics in the high-density vacancy expansion, and all even cumulants obey

$\gamma^{-1}$5

at long times, independently of the tumbling probability $\gamma^{-1}$6 (Bertrand et al., 2018). The full position distribution becomes Skellam-type, and the usual active signature is erased at leading order by single-file crowding (Bertrand et al., 2018). This shows that persistence can dominate transients yet disappear from the leading asymptotic transport law.

5. Generalized kinetics, anisotropy, and thermodynamic extensions

Several recent extensions alter the internal kinematics rather than the environment. In the inertial run-and-tumble particle,

$\gamma^{-1}$7

the two intrinsic times

$\gamma^{-1}$8

generate four distinct regimes with

$\gamma^{-1}$9

in successive windows, before the long-time diffusivity

$v$0

is recovered (Dutta et al., 2024). Inertia therefore creates transient universality classes absent from the overdamped RTP, including a random-acceleration-like regime with persistence exponent $v$1 (Dutta et al., 2024).

Anisotropic reorientation produces a different generalization. In run-and-tumble-turn dynamics, a two-dimensional active particle moves in a four-well angular potential that favors the Cartesian directions and turns by $v$2 between neighboring wells (Loewe et al., 2023). The long-time transport is governed by an effective tumble-turn time

$v$3

and once time and length are normalized by $v$4 and $v$5, the large-scale MSD, fourth moment, and non-Gaussian parameter become insensitive to the microscopic well width $v$6 (Loewe et al., 2023). The hydrodynamic limit is diffusive, but anisotropy survives in higher-gradient corrections, so the transition to diffusive dynamics precedes the transition to isotropic dynamics (Loewe et al., 2023).

A further extension replaces the two-state or four-state internal dynamics by a non-reciprocal three-state process with velocity states $v$7. The asymptotic drift and diffusion are

$v$8

where the six transition rates $v$9 need not satisfy reciprocity (Romo-Cruz et al., 13 Aug 2025). The model is asymptotically diffusive on the manifold

$\lambda$0

even though microscopic reversibility is absent (Romo-Cruz et al., 13 Aug 2025). This is a particularly sharp example of macroscopic diffusion emerging from intrinsically irreversible internal cycling.

The same logic has also been transplanted to phase dynamics. In a run-and-tumble Kuramoto model, each oscillator redraws its intrinsic frequency from $\lambda$1 at Poisson rate $\lambda$2, and the instability threshold of the incoherent state becomes

$\lambda$3

(Frydel, 2021). Here “run-and-tumble” no longer refers to spatial velocity reversal but to dynamic sampling of quenched disorder (Frydel, 2021).

Thermodynamic reformulations are beginning to appear as well. For a one-dimensional RTP in a confining potential and a thermal bath, the local entropy flux to the reservoir can be written exactly as

$\lambda$4

with $\lambda$5, and this can be used to construct an “Inverse Clausius Thermodynamics” with an inferred effective temperature field $\lambda$6 (Farago, 10 Sep 2025). This suggests that even in explicitly active settings, a local thermodynamic representation may be reconstructed from entropy exchange rather than postulated a priori.

6. Optimization, control, and engineered realizations

Run-and-tumble dynamics is also a design variable. In persistent chemotaxis, linearized performance has an optimal persistence for fixed base tumble frequency, but in the full $\lambda$7 plane there is a continuum of optimal solutions; finite tumble times that depend on persistence remove this degeneracy and provide a mechanism for selecting persistence from species-specific reorientation dynamics (Kirkegaard et al., 2017). Beyond linear response, optimal chemotactic strengths react strongly when the swimmer is moving in the wrong direction and only weakly when it already has even a slight projection up the gradient (Kirkegaard et al., 2017).

In confined one-dimensional RTPs, the full distribution can be steered exactly between nonequilibrium steady states by shortcut protocols. For the harmonic model

$\lambda$8

the exact control equations are

$\lambda$9

with the ansatz $\tau$0 (Frim et al., 2023). The paper shows that active-to-passive transformations can be substantially accelerated, whereas passive-to-active steering is intrinsically harder because the velocity sectors must separate spatially (Frim et al., 2023).

Engineered systems now realize run-and-tumble-like motion without explicitly programming a tumble state. Two mechanically coupled overdamped active Brownian robots linked by an off-center rigid rod exhibit sharp tumbles, exponentially distributed run times, and tunable tumbling frequency $\tau$1 (Paramanick et al., 3 Feb 2025). The reduced theory identifies metastable running manifolds and noise-activated escapes between them, so the run-and-tumble statistics emerge from mechanical coupling rather than a telegraphic switching rule (Paramanick et al., 3 Feb 2025). A related macroscopic realization is the soft robotic cell: a programmable active robot enclosed within a deformable membrane undergoes intermittent stop-and-go motion whose MSD is ballistic at short times and diffusive at long times, and whose effective run-and-tumble description uses

$\tau$2

(Mohapatra et al., 4 Jun 2026). Membrane softness then acts as a single control parameter tuning persistence, intermittency, and long-time transport (Mohapatra et al., 4 Jun 2026).

Taken together, these results suggest a broader interpretation. Run-and-tumble dynamics is not only a phenomenological model for microbial motility. It is also an effective language for systems in which persistence and reorientation are generated by internal mechanics, boundary-induced bistability, biochemical memory, or externally engineered control. A plausible implication is that the enduring utility of the run-and-tumble framework lies less in any particular microscopic mechanism than in its ability to organize persistent transport across scales, environments, and implementations.