---
title: 'Manna Sandpile Model: Dynamics & Universality'
url: https://www.emergentmind.com/topics/manna-sandpile-model
type: topic
---

# Manna Sandpile Model: Dynamics & Universality

The Manna sandpile model is a family of stochastic sandpile models in which unstable or active sites redistribute two particles randomly to nearest neighbours. Two formulations are central in the literature: the conserved Manna sandpile, also called a fixed-energy or conserved stochastic sandpile, where the total mass is strictly conserved and the control parameter is the density $\rho$; and the Abelian Manna Model, in which slow driving and boundary dissipation generate self-organized criticality. Across these formulations, the model is a canonical setting for absorbing-state phase transitions, avalanche scaling, hydrodynamic coarse graining, hyperuniformity, and universality questions involving directed percolation, conserved directed percolation, and interface depinning [1711.03793], [1106.0406].

## 1. Microscopic definitions and principal variants

In the one-dimensional conserved Manna sandpile, one considers a periodic chain of $L$ sites, each carrying an unbounded integer mass $m_i \ge 0$ or $n_X \ge 0$. A site is active iff $m_i>1$ or equivalently $n_X \ge 2$. The continuous-time dynamics is random sequential: an active site topples at rate $1$, loses two particles, and each particle is sent independently to one of the two nearest neighbours with probability $1/2$. The total mass $N=\sum_i m_i$ is strictly conserved, and the global density is $\rho=N/L$. In steady state for $\rho>\rho_c$, the order parameter is the activity,
\[
a(\rho)=\frac{1}{L}\sum_{i=1}^L \langle \hat a_i\rangle=\langle a_i\rangle,
\]
the mean density of active sites [1711.03793], [2211.14070].

The fixed-energy formulation exhibits an active-to-absorbing transition. For $\rho<\rho_c$ the density of active sites decays to zero and the system reaches an absorbing configuration; for $\rho>\rho_c$ activity persists in the infinite-size limit. Any configuration with $z_i\in\{0,1\}$ for all $i$ is absorbing, so the model has infinitely many absorbing states in the thermodynamic limit [1206.3121].

The Abelian Manna Model is the slowly driven counterpart. On a lattice with integer heights $z_i\ge 0$ or $z(x,y)\ge 0$, a site is unstable if $z_i>1$ or $z(x,y)\ge 2$. When the configuration is quiescent, one particle is added at a uniformly random site; if this creates an unstable site, an avalanche begins. Each unstable site topples by removing two particles and sending them independently to randomly chosen nearest neighbours. In open geometries, particles sent to virtual neighbours are lost at the boundary. Because the toppling operators commute, $\bigl[T_i,T_j\bigr]=0$, the final stable configuration and the total number of topplings do not depend on the order of updates [2311.03513], [1106.0406].

This distinction between conserved and slowly driven dynamics is structurally important. In the conserved model, density is the tuning parameter of an absorbing-state transition. In the Abelian model, slow driving and dissipation organize the system into a critical state with avalanche observables such as size $s$, duration $t$ or $T$, and area $a$ [1106.0406].

## 2. Hydrodynamic structure of the conserved model

A central result for conserved stochastic sandpiles is that their coarse-grained density dynamics closes through the activity. Under a local-steady-state or local equilibrium ansatz, the one-dimensional conserved Manna sandpile obeys
\[
\partial_t \rho(x,t)=\partial_x^2\,a\bigl(\rho(x,t)\bigr),
\]
or, in the presence of a weak biasing field $F$,
\[
\partial_t\rho(x,t)=\partial_x^2\,a\bigl(\rho(x,t)\bigr)-F\,\partial_x\,a\bigl(\rho(x,t)\bigr)\equiv -\partial_x J.
\]
Reading off the diffusive and drift currents yields
\[
J_{\rm diff}=-D(\rho)\,\partial_x \rho,\qquad D(\rho)=\frac{da}{d\rho}\equiv a'(\rho),
\]
and
\[
J_{\rm drift}=\chi(\rho)\,F,\qquad \chi(\rho)=a(\rho).
\]
Thus the bulk diffusion coefficient is the derivative of the activity, whereas the conductivity equals the activity itself [1711.03793], [2011.01173].

Macroscopic Fluctuation Theory and an additivity principle imply an equilibrium-like Einstein relation between transport and density fluctuations,
\[
\sigma^2(\rho)=\frac{\chi(\rho)}{D(\rho)}.
\]
For the conserved Manna sandpile this becomes
\[
\boxed{\sigma^2(\rho)=\frac{a(\rho)}{a'(\rho)}}.
\]
Here $\sigma^2(\rho)$ is the scaled variance of subsystem mass, defined by $\langle m^2\rangle-\langle m\rangle^2\approx v\,\sigma^2(\rho)$ for a macroscopic subsystem of volume $v$ [1711.03793].

The same hydrodynamic structure defines an equilibrium-like chemical potential and large-deviation principle. Writing $f(\rho)$ for the free-energy density, one has
\[
f''(\rho)=\frac{D(\rho)}{\chi(\rho)},\qquad \frac{d\mu}{d\rho}=\frac{a'(\rho)}{a(\rho)},
\]
so that
\[
\boxed{\mu(\rho)=\ln a(\rho)+\mathrm{const.}}
\]
The coarse-grained density $\hat \rho$ in a subvolume $v$ then satisfies
\[
P_v(m)\simeq \exp\bigl[-v\,h(\hat \rho)\bigr],\qquad h(\hat \rho)=f(\hat \rho)-\mu(\rho)\,\hat \rho.
\]
This gives the conserved Manna sandpile an equilibrium-like thermodynamic description at the level of subsystem mass fluctuations, despite its nonequilibrium microscopic dynamics [1711.03793].

Near the critical density $\rho_c$, with $\Delta=\rho-\rho_c\to 0^+$,
\[
a(\rho)\sim \Delta^\beta,\qquad \xi\sim \Delta^{-\nu_\perp}.
\]
Since $D=a'$, one obtains
\[
\sigma^2(\rho)\sim \Delta^{1-\delta},\qquad \delta=0,
\]
and the dynamic exponent
\[
\boxed{z=2+\frac{\beta-1}{\nu_\perp}}
\]
or equivalently $z=2-(1-\beta)/\nu_\perp$. For the one-dimensional unrestricted-height Manna sandpile, the quoted estimates are
\[
\rho_c\approx 0.95,\qquad \beta\approx 0.42,\qquad \nu_\perp\approx 1.81,\qquad z\approx 1.66,
\]
while inserting $\beta$ and $\nu_\perp$ into the scaling relation yields $z\approx 1.68$ [1711.03793].

## 3. Dynamic correlations, fluctuation suppression, and hyperuniformity

The conserved Manna sandpile also displays anomalous temporal correlations. Let $Q_i(T)=\int_0^T \mathcal{J}_i(t)\,dt$ be the cumulative bond current across bond $(i,i+1)$, with instantaneous current $\mathcal{J}_i(t)=dQ_i(t)/dt$. In the thermodynamic limit, for short times $1\ll T\ll L^2$,
\[
{\rm Var}\,Q_i(T)\sim T^\alpha,\qquad \alpha=\frac12-\mu,
\]
with
\[
\mu=\frac{\beta+1}{2\nu_\perp z}.
\]
Far from criticality, $\mu=0$ and $\alpha=1/2$; near criticality, $\mu>0$, so fluctuation growth is further suppressed. At very late times $T\gg L^2$, ordinary diffusive growth ${\rm Var}\,Q\sim T$ is recovered [2211.14070].

This temporal suppression is reflected in long-time correlations and low-frequency power spectra. The current autocorrelation decays as
\[
C_{\mathcal{J}\mathcal{J}}(t)\sim -A\,t^{-(3/2+\mu)},
\qquad A\propto \frac{a}{\sqrt{a'}}.
\]
Consequently, the bond-current power spectrum and subsystem-mass power spectrum behave as
\[
S_{\mathcal{J}}(f)\sim f^{1/2+\mu},\qquad S_M(f)\sim f^{-3/2+\mu},
\]
and their exponents satisfy the continuity-equation identity
\[
\psi_M+\psi_{\mathcal{J}}=2.
\]
The literature characterizes the case $\mu>0$ as dynamic hyperuniformity [2211.14070].

The fluctuation relations are unusually explicit. In the long-time, large-subsystem limit,
\[
\lim_{T\to\infty}\lim_{l\to\infty}\frac{\left\langle \left[\sum_{i=0}^{l-1}Q_i(T)\right]^2\right\rangle}{lT}=2\,a(\rho),
\]
so the integrated current variance per unit space-time equals $2a(\rho)$. Together with $D(\rho)=a'(\rho)$, this gives the exact Green-Kubo form
\[
\sigma^2(\rho)=\frac{\sigma_Q^2(\rho)}{2D(\rho)},\qquad \sigma_Q^2=2a(\rho),
\]
hence again
\[
\sigma^2(\rho)=\frac{a(\rho)}{a'(\rho)}.
\]
As $\Delta\to 0^+$, this vanishes linearly in $\Delta$, consistent with hyperuniform scaling of static density fluctuations [2211.14070].

Tagged-particle transport is likewise tied directly to activity. If $X(T)$ is the displacement of a tagged particle, then
\[
\sum_{\rm particles} X^2(T)=\sum_i Q_i(T),
\]
and for $T\gg L^2$ the self-diffusion coefficient defined by $\langle X^2(T)\rangle \simeq 2D_s T$ obeys
\[
D_s(\rho)=\frac{a(\rho)}{\rho}.
\]
Thus $D_s$ vanishes as $\Delta^\beta$ near criticality, while the bulk diffusion coefficient $D=a'$ diverges as $\Delta^{-(1-\beta)}$ [2211.14070]. A plausible implication is that bulk and tagged-particle transport probe different aspects of the same activity field.

## 4. Relaxation, anomalous transport, and front propagation

Long-wavelength density relaxation is diffusive only away from criticality. For a uniformly supercritical background with finite correlation length $\xi$, linearizing the nonlinear diffusion equation gives
\[
\omega(k)=D(\rho)\,k^2,\qquad \tau_R(k)\simeq \frac{1}{D(\rho)\,k^2}.
\]
In a finite system of size $L$, the slowest mode decays on a time scale of order $L^2/D(\rho)$. Simulations of step, wedge, and Gaussian initial profiles collapse under the diffusive scaling $\rho(X,t)=g(X/L,t/L^2)$ [2011.01173].

Near criticality, simple diffusion breaks down. Since
\[
a(\rho)\sim (\rho-\rho_c)^\beta,\qquad \beta<1,
\]
the bulk diffusivity diverges:
\[
D(\rho)=a'(\rho)\sim (\rho-\rho_c)^{\beta-1}\to \infty\qquad (\rho\to \rho_c^+).
\]
Once the perturbation wavelength $k^{-1}$ becomes comparable to the correlation length $\xi\sim (\rho-\rho_c)^{-\nu_\perp}$, transport is anomalous. Finite-size or finite-wavelength scaling then yields
\[
\tau_R(k)\sim k^{-z},\qquad z=2-\frac{1-\beta}{\nu_\perp}<2,
\]
which is the same relation as $z=2+(\beta-1)/\nu_\perp$ [2011.01173].

On a critical background, a localized excess-density perturbation spreads self-similarly. With $\Delta=\rho-\rho_c$ and $a(\Delta)\simeq C\Delta^\beta$, the asymptotic equation is
\[
\frac{\partial \Delta}{\partial t}
=
C\,\frac{\partial^2}{\partial X^2}\bigl[\Delta^\beta\bigr].
\]
A self-similar solution
\[
\Delta(X,t)=(Ct)^{-\omega}\,\mathcal{G}\!\left(\frac{X}{(Ct)^\omega}\right)
\]
exists with
\[
\omega=\frac{1}{1+\beta}>\frac12,
\]
and
\[
\mathcal{G}(y)=\bigl[B_0+B\,y^2\bigr]^{-1/(1-\beta)}.
\]
Hence the perturbation width grows anomalously,
\[
\sigma(t)\sim t^\omega.
\]
Monte Carlo data for Gaussian and delta-like initial pulses collapse onto the predicted scaling function over several decades in $\Delta$ [2011.01173].

The same hydrodynamic equation describes invasion fronts between subcritical and supercritical regions. Numerically integrating $\partial_t\rho=\partial_X^2 a(\rho)$ with $a(\rho<\rho_c)=0$ reproduces a moving front, the boundary-layer law
\[
\rho(X)-\rho_c\sim (\mathrm{distance})^{3/2},
\]
and the front position
\[
X_*(t)\sim t^{1/2}.
\]
For global density $\bar \rho<\rho_c$ on a finite ring, invaded regions freeze at $\rho_c$; for $\bar \rho>\rho_c$, the whole system becomes active [2011.01173].

## 5. Avalanches, finite-size scaling, and substrate correlations in the Abelian Manna Model

The slowly driven Abelian Manna Model is a standard self-organized critical sandpile. Its primary avalanche observables are the size $s$, the duration $t$ or $T$, and the area $a$. The standard finite-size-scaling ansatz is
\[
P_s(s;L)=a_s\,s^{-\tau_s}G_s(s/s_c),\qquad
P_t(t;L)=a_t\,t^{-\tau_t}G_t(t/t_c),\qquad
P_a(a;L)=a_a\,a^{-\tau_a}G_a(a/a_c),
\]
with cutoffs
\[
s_c\sim L^D,\qquad t_c\sim L^z,\qquad a_c\sim L^{D_a}.
\]
For $n+1>\tau_x$,
\[
\langle x^n\rangle(L)\sim L^{D_x(n+1-\tau_x)},\qquad x\in\{s,t,a\}.
\]
Moment analysis across four one-dimensional and eight two-dimensional lattices yields lattice-independent exponents within errors, supporting universality across different geometries [1106.0406].

The quoted exponent estimates are, in $d=1$,
\[
D=2.253(14),\quad \tau_s=1.112(6),\quad
z=1.445(10),\quad \alpha=1.18(2),\quad
D_a=0.998(3),\quad \tau_a=1.259(11),
\]
with
\[
\Sigma\equiv D(\tau_s-1)=z(\alpha-1)=D_a(\tau_a-1)=0.26(2).
\]
In $d=2$,
\[
D=2.750(6),\quad \tau_s=1.273(2),\quad
z=1.532(8),\quad \alpha=1.4896(96),\quad
D_a=1.995(3),\quad \tau_a=1.382(3),
\]
and
\[
\Sigma=0.761(13).
\]
Particle conservation in the bulk implies the exact scaling relation
\[
D(2-\tau_s)=2,
\]
equivalently $\langle s\rangle\sim L^2$, and the numerical results satisfy the narrow-joint-distribution identity
\[
\Sigma=D(\tau_s-1)=z(\alpha-1)=D_a(\tau_a-1)
\]
within errors [1106.0406].

Corrections to scaling are notable in the Manna universality class. In the moment fits, leading corrections were empirically found to be $N^{-1/4}$ and $N^{-1/2}$, and with these terms the fits reached high goodness-of-fit $(q>0.9)$ for all non-trivial moments $n\ge 2$. Moment ratios
\[
g_{x,n}=\frac{\langle x^n\rangle\,\langle x\rangle^{\,n-2}}{\langle x^2\rangle^{\,n-1}}
\]
are also lattice-independent, especially in $d=2$ [1106.0406].

Two-dimensional studies of the Abelian Manna Model further show that weak substrate correlations matter specifically for the largest avalanches. After each non-zero avalanche, one may destroy correlations by shuffling configurations across an ensemble of replicas in a way that preserves the one-point density profile but destroys two-point correlations between different $x$-layers. In the unshuffled model, the best estimates are
\[
\tau=1.276(2),\quad D=2.763(8),\quad
\alpha=1.497(8),\quad z=1.542(5).
\]
In the fully shuffled model, moment analysis yields instead
\[
\tau=1.017(6),\quad D=2.028(6),\quad
\alpha=1.138(26),\quad z=1.050(10)
\]
for the large-$s$ tail. However, the distributions show two scaling regimes: for small avalanches up to
\[
s_\times(L)\sim L^{0.72},
\]
the exponent remains $\tau_{\rm small}\approx 1.276$, while for large avalanches the tail crosses to $\tau_{\rm large}=1.017$ with cutoff $s_c(L)\propto L^{2.028}$. No single power law plus cutoff describes the entire range [2311.03513].

This establishes that destroying local substrate correlations does not alter small and intermediate avalanches, but changes the scaling of the largest, system-wide events. The result is a crossover and a breakdown of simple finite-size scaling in the decorrelated model [2311.03513].

## 6. Effective theories, critical densities, and universality issues

Several complementary descriptions of the Manna model have been developed. For critical absorbing states in one dimension, inverse Ising inference reconstructs an effective equilibrium Hamiltonian
\[
H\bigl(\{n_i\}\bigr)
=
\frac12\sum_{i,j=1}^L J(|i-j|)\,n_i n_j-\mu\sum_{i=1}^L n_i.
\]
Below $\rho_c$, the inferred interaction has a short-range repulsive core and a long-range attractive tail. At criticality, the interaction becomes purely repulsive,
\[
J(r)\sim J_0\,r^{-\alpha},
\]
with the conjectured exact exponent
\[
\alpha=\frac12.
\]
Standard Fourier analysis then gives the window-variance exponent
\[
\lambda=d+\alpha=\frac32
\]
in $d=1$, and the associated Gaussian field theory is
\[
\mathcal{H}[\phi]=\frac12\int \frac{dk}{2\pi}\,|k|^{-1/2}|\phi(k)|^2,
\]
implying
\[
\langle \phi(x)\phi(0)\rangle\sim |x|^{-3/2}.
\]
This description links critical absorbing states to hyperuniformity and to quenched-EW-type depinning models [1809.08592].

At the continuum level, the stochastic field theory of the Manna or C-DP class admits an exact mapping to an overdamped elastic interface in short-range correlated quenched disorder. After a change of variables, the interface height $u(x,t)$ obeys
\[
\partial_t u(x,t)
=
(\nabla^2-m^2)u(x,t)+F\bigl(u(x,t),x\bigr)+f(x)+F_{\rm ret}(x,t),
\]
where $F_{\rm ret}$ is a retarded memory term. The mapping shows that this memory term is RG-irrelevant, so the universal large-scale dynamics is that of the quenched Edwards-Wilkinson depinning model. In this formulation, the sandpile order parameter maps to the mean interface velocity, and avalanche size is the total interface advance [1410.1930].

The model is also central to the distinction among several density notions. In a quenched version of the Manna sandpile on directed environments, the stationary density and transition density are not equal:
\[
\rho_s=1-\frac12 f_r,
\]
while $\rho_{tr}$ is the unique positive solution of
\[
\rho-\frac12(1+e^{-2\rho})
=
\frac{1-f_r}{2}(1+e^{-2\rho})^2.
\]
In the same setting,
\[
\rho_{th}(\nu)=\rho_{tr}(\nu),
\]
so $\rho_s\neq \rho_{tr}=\rho_{th}$. For the annealed Manna model, the rigorous bound $\rho_{tr}\ge 1/4$ is quoted, simulations suggest $\rho_{tr}\approx 0.9489$, $\rho_{th}=1$, and it is conjectured that $\rho_s\approx \rho_{tr}$ [1211.4760].

Universality remains contested in part of the literature. One numerical study of selected $(1+1)$-dimensional fixed-energy sandpiles, using natural homogeneous initial states, reported
\[
\alpha\approx 0.159,\quad \beta\approx 0.28,\quad \nu_\perp\approx 1.095,\quad \nu_\parallel\approx 1.74,\quad z\approx 1.58
\]
and argued that the critical behavior converges to directed percolation after very long times [1206.3121]. By contrast, the hydrodynamic treatment of conserved stochastic sandpiles derives the scaling relation
\[
z=2+\frac{\beta-1}{\nu_\perp},
\]
which ordinary directed percolation in $1D$ does not satisfy: with its quoted exponents $\beta\approx 0.28$, $\nu_\perp\approx 1.10$, the relation would predict $z\approx 1.35$, not the measured $z\approx 1.58$ [1711.03793]. Additional numerical evidence for a distinct Manna class comes from the random rotational sandpile model, where a continuous crossover culminates at the maximally disordered point $p=0.5$ with exponents
\[
\tau_s(0.5)=1.286(9),\qquad D_s(0.5)=2.71(2),\qquad
H(0.5)=0.21\pm 0.01,\qquad \chi(0.5)=0.70\pm 0.02,
\]
matching the best-known Manna values within errors [1501.01748].

Taken together, these results show that the Manna sandpile model is not only a stochastic redistribution process but also a nexus connecting absorbing-state criticality, self-organized criticality, hydrodynamic fluctuation theory, hyperuniformity, and depinning-like field theories. The principal unresolved issue is not whether the model is critical, but how its conserved dynamics should be situated relative to directed percolation, conserved directed percolation, and quenched Edwards-Wilkinson universality descriptions.

Source: https://www.emergentmind.com/topics/manna-sandpile-model