Zero-Range Process: Dynamics and Condensation

• Zero-range process is a conservative interacting particle system where jump rates depend solely on the number of particles at a site, enabling exact stationary product measures.
• Its flexible rate functions allow rigorous analysis of hydrodynamic limits, condensation transitions, and non-equilibrium phenomena under varied conditions.
• Extensions to boundary-driven and multispecies models highlight the framework's broad relevance in statistical mechanics and complex transport theory.

A zero-range process (ZRP) is a conservative interacting particle system in which the jump rate out of a site depends only on the number of particles at that departure site. In its standard form, particles move on a lattice or graph, the local rate function $g$ encodes the interaction, and the resulting dynamics admit a remarkably broad range of behaviors: exact stationary product measures, diffusive or hyperbolic hydrodynamic limits, condensation transitions, metastable condensate motion, non-equilibrium stationary states, and integrable multispecies generalizations (Bahadoran et al., 2020, Balázs et al., 2020, Armendáriz et al., 2015).

1. Definition and core structure

In a general ZRP, a configuration is an occupation field $\eta=(\eta_x)_{x\in\Lambda}$ with $\eta_x\in\mathbb N_0$, and one particle moves from $x$ to $y$ at a rate determined by $g(\eta_x)$ and a jump kernel $p(x,y)$. A standard generator, appearing in several formulations, is

$$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$

where $\eta^{x,y}$ is the configuration after moving one particle from $x$ to $\eta=(\eta_x)_{x\in\Lambda}$0 (Balázs et al., 2020). This framework includes finite lattices, infinite lattices such as $\eta=(\eta_x)_{x\in\Lambda}$1, complete graphs, open systems with reservoirs, long-jump kernels, and quenched random environments (Shargel et al., 2010, Bernardin et al., 2022).

A central structural assumption is often $\eta=(\eta_x)_{x\in\Lambda}$2, together with monotonicity of $\eta=(\eta_x)_{x\in\Lambda}$3. When $\eta=(\eta_x)_{x\in\Lambda}$4 is nondecreasing, the process is attractive, so monotone couplings and order-preservation become available; this is a key ingredient in hydrodynamic limits, strong local equilibrium, and coupling-based mixing arguments (Bahadoran et al., 2020, Marahrens et al., 2024). The rate function can be linear, as in $\eta=(\eta_x)_{x\in\Lambda}$5; constant on occupied sites, as in $\eta=(\eta_x)_{x\in\Lambda}$6; decreasing, as in $\eta=(\eta_x)_{x\in\Lambda}$7; or even system-size dependent, such as

$\eta=(\eta_x)_{x\in\Lambda}$8

which produces a discontinuous condensation transition (Armendáriz et al., 2015, Chleboun et al., 2014). There are also constructions with rapidly growing, even superlinear, rates on $\eta=(\eta_x)_{x\in\Lambda}$9, extending the standard bounded-increment regime (Andjel et al., 2020). This range of choices is one reason the ZRP functions less as a single model than as a flexible class of exactly analyzable interacting particle systems.

2. Stationary product measures, fugacity, and ensembles

A defining feature of many ZRPs is the existence of stationary product measures. Writing

$\eta_x\in\mathbb N_0$0

the one-site grand-canonical law is typically

$\eta_x\in\mathbb N_0$1

with product measure $\eta_x\in\mathbb N_0$2 (Balázs et al., 2020). The mean density is

$\eta_x\in\mathbb N_0$3

and when $\eta_x\in\mathbb N_0$4 is invertible, its inverse $\eta_x\in\mathbb N_0$5 maps density to activity (Bernardin et al., 2022). A recurrent identity is

$\eta_x\in\mathbb N_0$6

which is the microscopic origin of the nonlinear flux $\eta_x\in\mathbb N_0$7 in hydrodynamic equations (Balázs et al., 2020).

This structure persists in several nonstandard settings. In right-biased disordered ZRPs on $\eta_x\in\mathbb N_0$8, the invariant product measure has site-dependent fugacity $\eta_x\in\mathbb N_0$9, yielding

$x$0

with a maximal invariant measure at the critical fugacity $x$1 (Bahadoran et al., 2020). On randomly oriented bistochastic graphs, the usual product distribution remains stationary because the divergence-free orientation balances incoming and outgoing mass in the quenched environment (Balázs et al., 2020). In boundary-driven long-jump ZRPs, the non-equilibrium stationary state is again product, but with site-dependent fugacities $x$2 solving a linear traffic equation (Bernardin et al., 2022).

The canonical ensemble is obtained by conditioning the product measure on fixed total mass, and ensemble equivalence or non-equivalence becomes a major question in condensation theory. In finite-capacity models, the single-site partition function is truncated,

$x$3

and the paper on finite compartments emphasizes that grand-canonical analysis remains exact even in condensed phases, unlike the infinite-capacity case where grand-canonical descriptions may fail above the critical density (Ryabov, 2013). This suggests that the product-measure formalism is robust, but the thermodynamic interpretation of those products depends sharply on geometry, capacity constraints, and the asymptotics of $x$4.

3. Hydrodynamic and hydrostatic limits

The macroscopic evolution of ZRPs is controlled by the same activity function $x$5 or $x$6 that appears in equilibrium. For the symmetric ZRP on the discrete torus $x$7 in dimensions $x$8, under diffusive scaling the empirical density converges to the nonlinear diffusion

$x$9

with $y$0 determined by the second moment of the jump kernel (Marahrens et al., 2024). A notable recent development is a quantitative consistency–stability method that yields convergence rates, uniform in time, in a modulated Monge–Kantorovich distance and in relative entropy, while avoiding block estimates (Marahrens et al., 2024).

Microscopic asymmetry does not by itself force hyperbolic behavior. On a two-lane torus with randomly oriented, divergence-free directed edges, a totally asymmetric and non-reversible ZRP still has a diffusive quenched hydrodynamic limit,

$y$1

because the local zero-flux constraint removes macroscopic drift (Balázs et al., 2020). The main technical novelty there is an environment-adapted tile decomposition and a second-order local equilibrium that neutralizes the usual non-gradient obstruction in entropy methods (Balázs et al., 2020).

By contrast, in right-biased disordered one-dimensional ZRPs the natural scale is hyperbolic, and the macroscopic law is a scalar conservation equation

$y$2

with truncated flux $y$3 above the critical density $y$4 (Bahadoran et al., 2020). That theory extends beyond subcritical profiles: it includes hydrodynamics for supercritical data, pseudo-equilibria above $y$5, and strong local equilibrium results showing convergence to the critical measure in supercritical regions (Bahadoran et al., 2020).

Open and nonlocal settings produce further variants. For the long-jump boundary-driven ZRP on $y$6, the hydrostatic profile is governed by regional fractional operators, and the stationary current satisfies a nonlocal Fick law in terms of the fugacity profile $y$7 (Bernardin et al., 2022). For the symmetric finite-segment ZRP with slow boundary creation at site $y$8 and slow annihilation at site $y$9, the hydrostatic limit is explicit, and the proposed diffusive hydrodynamic equation is

$g(\eta_x)$0

with boundary conditions that depend on $g(\eta_x)$1: Robin-type when $g(\eta_x)$2, and Neumann when $g(\eta_x)$3 (Frómeta et al., 2020).

4. Condensation and its mechanisms

Condensation is the phenomenon whereby, above a critical density, a macroscopic fraction of the mass concentrates on a vanishing fraction of sites. In the classical homogeneous setting with decreasing rates

$g(\eta_x)$4

the one-site weights satisfy $g(\eta_x)$5, the critical density is finite for $g(\eta_x)$6, and in the supercritical regime a single site carries the excess mass $g(\eta_x)$7 while the background remains distributed as the critical product measure (Armendáriz et al., 2015). This is the canonical condensing ZRP on which metastability theory has been built.

The condensation mechanism, however, is not confined to decreasing rates. In a generalized conserved interacting ZRP with pairwise updates, condensation appears for a range of transfer bias $g(\eta_x)$8, but the stationary measure does not factorize because the rates depend on both occupancies of the interacting pair (Khaleque et al., 2016). In short-range dynamics the condensate-phase bump has universal scaling exponent $g(\eta_x)$9, whereas in long-range mean-field dynamics the exponent is non-universal and depends on $p(x,y)$0 (Khaleque et al., 2016). This shows that condensation can survive even after the standard product-measure structure is broken.

Disorder can either create condensation or suppress it. In randomly perturbed ZRPs with

$p(x,y)$1

the homogeneous condensation criterion is altered drastically: with any $p(x,y)$2, condensation in the thermodynamic limit survives only for $p(x,y)$3, while for $p(x,y)$4 the quenched critical density becomes infinite and condensation disappears (Molino et al., 2012). By contrast, in a ZRP whose one-particle stationary distribution is multifractal, condensation is induced by quenched inhomogeneity even though the microscopic hopping is the “pure chipping” case $p(x,y)$5; the non-condensed mass scales algebraically as

$p(x,y)$6

with exponent determined by the disorder strength (Miki, 2016).

A particularly sharp modification is the fast-rate model in which one occupation level $p(x,y)$7 carries a diverging rate $p(x,y)$8, while the rest of the rates are constant. In that model the critical density is finite, local marginals converge to the critical product law above $p(x,y)$9, and the condensed phase depends on the scaling $$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$0: for $$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$1, the excess mass splits into many clusters of typical size

$$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$2

whereas for $$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$3 it collapses to a single condensate site (Jatuviriyapornchai et al., 2024). Finite capacities provide yet another mechanism: with site capacities $$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$4, condensation can coexist with exact grand-canonical analysis and exhibit self-blocking and glassy relaxation (Ryabov, 2013). Taken together, these results show that condensation is not equivalent to one specific tail condition on $$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$5; it can arise, disappear, or change structure under disorder, capacity constraints, or even a single diverging fast rate.

5. Metastability, coarsening, and mixing

Once condensation occurs, the dominant dynamical questions concern motion of the condensate, coarsening toward the condensed state, and the time to equilibrium. For the one-dimensional symmetric condensing ZRP on a ring at fixed supercritical density, the condensate location is metastable on the time scale

$$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$6

and the rescaled condensate location converges to a symmetric pure-jump Lévy process on the unit torus, with jump kernel determined by capacities of a single random walk on the lattice (Armendáriz et al., 2015). The same analysis yields equilibration estimates within metastable wells, capacity asymptotics, and a uniform bound on well-exit rates (Armendáriz et al., 2015).

In a size-dependent ZRP with

$$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$7

the free-energy landscape is explicitly metastable. The canonical free energy has competing fluid and condensed minima, and within the condensed region there is a dynamical transition line

$$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$8

separating two distinct condensate-relocation mechanisms: evaporation–reformation through a fluid saddle, and splitting–coalescence through a two-condensate saddle (Chleboun et al., 2014). The associated transition times satisfy Arrhenius-type asymptotics governed by the relevant large-deviation barriers (Chleboun et al., 2014).

Before equilibrium is reached, condensing ZRPs exhibit coarsening. On the complete graph, for

$$(Lf)(\eta)=\sum_{x}\sum_{y} p(x,y)\,g(\eta_x)\big[f(\eta^{x,y})-f(\eta)\big],$$9

the coarsening scale is $\eta^{x,y}$0, and both at criticality and in the condensed phase finite-time corrections are essential to the correct scaling description (Godreche et al., 2016). At criticality the scaling function develops a front at

$\eta^{x,y}$1

with Gaussian smoothing at finite times; in the condensed phase the excess mass sits in a scaling bump whose amplitude approaches a selected value only slowly (Godreche et al., 2016).

Mixing theory reveals a parallel picture in mean-field non-condensing models. For the unit-rate ZRP on the complete graph, with density $\eta^{x,y}$2, the worst-case total-variation distance exhibits cutoff at

$\eta^{x,y}$3

and more generally the mixing time from a given initial state depends explicitly on the largest initial occupancies (Merle et al., 2018). For mean-field ZRPs with unbounded monotone rates $\eta^{x,y}$4, bounded density, and

$\eta^{x,y}$5

the mixing time is asymptotic to $\eta^{x,y}$6, where $\eta^{x,y}$7 is the largest initial height, and the chain has cutoff at $\eta^{x,y}$8 when $\eta^{x,y}$9; the Poincaré constant is $x$0 in mean field and compares to the underlying graph spectral gap on general geometries (Tran, 2021).

6. Boundary-driven, multispecies, and related extensions

Several extensions show how far the ZRP framework can be pushed without losing analytic control. In open one-dimensional systems with injection at the left boundary, hopping to the right, decay in the bulk, and absorption at the target site, the arrival process can be computed exactly in the bulk-dynamics case. If each particle hops with rate $x$1 and decays with rate $x$2, then the asymptotic arrival rate at site $x$3 is

$x$4

and the arrival counting process is Poisson in the independent-particle regime (Shargel et al., 2010). This connects ZRPs to first-passage and queueing problems as directly as to equilibrium statistical mechanics.

Multispecies and integrable variants replace scalar occupation fields by species-labeled multiplicities. In the $x$5-species totally asymmetric ZRP on a periodic chain, smaller species have priority to hop, the process is realized as the image of an $x$6-line process with uniform steady state, and the stationary weights admit a matrix product formula

$x$7

where the $x$8 are built from a $x$9 oscillator algebra and combinatorial $\eta=(\eta_x)_{x\in\Lambda}$00-matrices of $\eta=(\eta_x)_{x\in\Lambda}$01 at $\eta=(\eta_x)_{x\in\Lambda}$02 (Kuniba et al., 2015). In a different totally asymmetric special limit with $\eta=(\eta_x)_{x\in\Lambda}$03 for $\eta=(\eta_x)_{x\in\Lambda}$04, algebraic Bethe ansatz yields explicit transition probabilities that also serve as exponential generating functions for directed random walks on multi-dimensional simplicial lattices (Bogoliubov et al., 2017).

The ZRP also appears as a dynamic environment. In a nearest-neighbor symmetric ZRP on $\eta=(\eta_x)_{x\in\Lambda}$05 started from equilibrium, if particles on the negative axis are initially infected and healthy particles become infected upon sharing a site with an infected one, then the infection front

$\eta=(\eta_x)_{x\in\Lambda}$06

has positive and finite asymptotic speed bounds (Baldasso et al., 2018). A central ingredient is a space-time decoupling estimate derived by sprinkling, which controls correlations between trajectory functionals supported in sufficiently separated boxes (Baldasso et al., 2018). This suggests that ZRPs are not only models of transport and aggregation, but also natural interacting backgrounds for reaction, spread, and propagation phenomena.