Oslo Model: One-Dimensional Sandpile Dynamics

• The Oslo model is a one-dimensional stochastic sandpile automaton characterized by random local thresholds that drive avalanches and foster self-organized criticality.
• It exhibits hyperuniform behavior with suppressed long-range fluctuations, where spatial stress variance scales as L^(1/2) and temporal avalanche correlations lead to deterministic extensive parts.
• The model’s scaling structure, with critical exponents matching those of the quenched Edwards–Wilkinson universality class, provides insights into complex SOC phenomena.

The Oslo model is a one-dimensional stochastic sandpile, or rice-pile, automaton in which local thresholds are renewed after every toppling. In its standard boundary-driven form, stress is added at one edge and the system relaxes through avalanches to a statistically stationary critical state. The model is important because it combines self-organized criticality, nontrivial stationary correlations, hyperuniformity, and a close but not microscopically exact correspondence with the quenched Edwards–Wilkinson (qEW) depinning universality class (Grassberger et al., 2016). Subsequent work has shown that the model is hyperuniform not only in its stationary spatial stress profile but also in the temporal sequence of avalanche sizes, and more recently an explicit formula for the stationary state has been derived as a constrained path sum in discrete space-time (Garcia-Millan et al., 2017, Lallemant et al., 8 Aug 2025).

1. Definition and dynamical rules

The one-dimensional Oslo model is defined on sites $i=1,\dots,L$. Each site carries an integer local stress $z_i\ge 0$ and a local threshold $z_i^*$, chosen randomly to be either $2$ or $3$. A site is stable if $z_i<z_i^*$ and unstable if $z_i\ge z_i^*$. For a bulk site $1

$z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$

At the boundaries only the existing neighbor is incremented; the other unit is lost at the edge. After each toppling, the threshold at the toppled site is reset randomly and independently to $2$ or $z_i\ge 0$0. This reset is the source of stochasticity and of the model’s effective stickiness (Grassberger et al., 2016).

The original self-organized-critical version is boundary-driven with open boundaries. One increases $z_i\ge 0$1 by one unit, then lets the system relax completely through a whole avalanche before adding again. The same framework admits bulk driving, where stress is added at a random interior site, and a fixed-energy sandpile (FES) version with periodic boundaries, in which no stress is lost and activity is studied as a function of mean stress. In the driven setting, the stationary state is the statistically steady distribution of stable recurrent configurations reached after long time under repeated drive-and-relax cycles; in the FES language, the analogous object is the natural critical state, the correlated ensemble of absorbing or near-critical configurations appropriate to the critical point (Grassberger et al., 2016).

2. Stationary critical state and hyperuniform organization

A central result for the one-dimensional model is that the stationary critical state is hyperuniform. In the spatial setting considered for the Oslo model, the relevant observable is the total stress $z_i\ge 0$2 in a block of length $z_i\ge 0$3. The large-scale density fluctuations are suppressed by negative long-range correlations, and the block-stress variance scales as

$z_i\ge 0$4

with

$z_i\ge 0$5

Using $z_i\ge 0$6, this gives

$z_i\ge 0$7

and the numerical data are consistent with

$z_i\ge 0$8

This relation between the hyperuniformity exponent and the correlation-length exponent is one of the model’s key structural results (Grassberger et al., 2016).

The same suppression of fluctuations appears in the temporal domain. For the boundary-driven one-dimensional model, if

$z_i\ge 0$9

is the sum of $z_i^*$0 consecutive avalanche sizes, then the variance scales as

$z_i^*$1

with the minimal hyperuniform exponent

$z_i^*$2

Equivalently, the sample-mean variance obeys

$z_i^*$3

The exact decomposition

$z_i^*$4

shows that the extensive part is deterministic and the residual $z_i^*$5 is bounded. This means that successive avalanche sizes are anticorrelated so precisely that the usual $z_i^*$6 variance of an average of independent samples is replaced by $z_i^*$7. The same work also shows that this temporal hyperuniformity is tied to fixed boundary drive and disappears for bulk driving, where the fluctuations are asymptotically Poisson-like (Garcia-Millan et al., 2017).

3. Critical exponents and scaling structure

The model’s critical behavior is conventionally expressed in terms of the distance from critical density,

$z_i^*$8

At criticality, finite-size scaling gives

$z_i^*$9

From large-system data,

$2$0

consistent with

$2$1

The density depletion near an open boundary,

$2$2

is again consistent with $2$3 (Grassberger et al., 2016).

For boundary-driven avalanches, the scaling ansatz is

$2$4

Because $2$5 exactly for boundary drive, one has

$2$6

The numerical estimate

$2$7

is consistent with

$2$8

Moment analysis gives

$2$9

For temporal spreading,

$3$0

with data supporting

$3$1

In the FES setting, the stationary active-site density obeys

$3$2

and the scaling relation

$3$3

yields

$3$4

Critical relaxation uses

$3$5

with

$3$6

The same analysis gives $3$7 for subcritical FES seed avalanches (Grassberger et al., 2016).

Quantity	Value	Context
$3$8	$3$9	Correlation length
$z_i<z_i^*$0	$z_i<z_i^*$1	Avalanche fractal dimension
$z_i<z_i^*$2	$z_i<z_i^*$3	Boundary-driven avalanche size
$z_i<z_i^*$4	$z_i<z_i^*$5	Dynamical exponent
$z_i<z_i^*$6	$z_i<z_i^*$7	Survival probability
$z_i<z_i^*$8	$z_i<z_i^*$9	Surviving-avalanche activity
$z_i\ge z_i^*$0	$z_i\ge z_i^*$1	FES order parameter
$z_i\ge z_i^*$2	$z_i\ge z_i^*$3	FES critical relaxation
$z_i\ge z_i^*$4	$z_i\ge z_i^*$5	Temporal correlation length
$z_i\ge z_i^*$6	$z_i\ge z_i^*$7	Spatial hyperuniformity exponent

4. Interface representation and qEW correspondence

A standard representation introduces an interface height $z_i\ge z_i^*$8, defined as the number of topplings at site $z_i\ge z_i^*$9 up to time $1

$1

and therefore

$1

The toppling update is

$1

with

$1

This gives the effective evolution

$1

By contrast, the standard qEW equation is

$1

The Oslo model is therefore not microscopically identical to standard qEW, because the effective noise depends on the neighboring heights as well as on $1$z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$0. Nonetheless, its scaling behavior is consistent with the qEW universality class (Grassberger et al., 2016).

The interface roughness is superrough globally. With

$z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$1

the critical scaling is of Family–Vicsek type, with

$z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$2

Because $z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$3, the local roughness shows anomalous scaling. The local-roughness analysis gives

$z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$4

and therefore

$z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$5

Both interface choices, $z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$6 and $z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$7, display the same anomalous scaling (Grassberger et al., 2016).

A later anchored-interface analysis recast the model in a broader family of one-dimensional advected systems with an absorbing boundary. In that formulation, the Oslo field shares a roughness exponent $z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$8 with anchored advected interfaces, while the dynamic exponent remains distinct, with a conjectured value $z_i \to z_i-2,\qquad z_{i\pm 1}\to z_{i\pm1}+1.$9. Because the Oslo height is the gradient of a depinning displacement field in that mapping, the same analysis infers

$2$0

for the driven elastic string at depinning (Shapira et al., 2023). The notation here differs from the spatial hyperuniformity exponent $2$1 used for block-stress fluctuations.

5. Numerical control, transients, and the natural critical state

A major part of the modern understanding of the Oslo model comes from recognizing that many earlier simulations were dominated by finite-size and transient effects. In open SOC systems, boundaries strongly affect profiles and avalanche statistics unless $2$2 is very large. In bulk-driven open systems, the random distance from the addition site to the boundary introduces an extra length scale, making corrections much worse than in boundary driving. In the FES setting, a finite maximum observation time $2$3 is especially dangerous, because critical and hyperuniform correlations then develop only up to scales of order

$2$4

At larger scales the configuration still resembles the initial state rather than the natural critical state, biasing $2$5, $2$6, $2$7, and $2$8 (Grassberger et al., 2016).

To avoid this, large-scale studies use periodic initial configurations with density very close to $2$9 and unusually small fluctuations, often built from repeated words of 1s and 2s chosen to be as uniform as possible. For subcritical FES seed avalanches, a useful protocol is to trigger a new avalanche by declaring a site with $z_i\ge 0$00 unstable, because this does not alter the background natural critical state itself. For supercritical FES, the system is evolved until a stationary active state is reached. The paper comparing initialization schemes found that restarting from a periodic near-critical background, labeled “scheme A,” is more reliable than reusing the previous final state with many newly declared unstable sites, labeled “scheme B,” because the latter produces pronounced transient distortions (Grassberger et al., 2016).

These methodological changes are not secondary. They are tied directly to hyperuniformity: if the stationary state suppresses long-wavelength fluctuations, then random initial conditions require very long transients before the correct negative correlations are built. The use of nearly hyperuniform initial states made it possible to simulate driven systems with $z_i\ge 0$01 and many FES studies with $z_i\ge 0$02 (Grassberger et al., 2016).

6. Universality, exact stationary state, and later developments

The universality-class assignment of the Oslo model has been a recurrent issue. Comparative work on the Abelian Manna model used the Oslo model as a benchmark for robust SOC avalanche scaling and reported closely matching avalanche exponents in one and two dimensions, which strengthened the view that the two models might share a common SOC universality class (Huynh et al., 2011). A later reanalysis of the one-dimensional Oslo model argued instead that, once finite-size and transient effects are controlled, the model is not in the directed-percolation/Manna class and is better understood as belonging to the qEW universality class (Grassberger et al., 2016). The distinction is not about the existence of avalanche scaling itself, but about which long-wavelength field theory captures the asymptotic behavior.

A further development concerns the stationary state itself. Recent work derived an explicit stationary-state expression for the one-dimensional Oslo model by using different representations of configurations and trajectories, identifying invariants attached to each configuration, and summing over all dynamically allowed paths leading to a given configuration under the corresponding invariant constraints. The resulting stationary probability is a weighted sum over paths, with weights determined by the threshold probabilities $z_i\ge 0$03 and $z_i\ge 0$04, and the construction is interpreted exactly as a parametrized path-integral formulation in discrete space-time (Lallemant et al., 8 Aug 2025). This provides an explicit description of the stationary measure itself, rather than only of its consequences for avalanche and roughness observables.

Taken together, these results define the Oslo model as a particularly well-controlled representative of one-dimensional driven criticality: a stochastic sandpile with exact boundary-driven conservation laws, a hyperuniform stationary state, strong temporal anticorrelations in avalanche sizes, and interface scaling consistent with qEW depinning even though its microscopic noise structure is more complicated than that of the standard qEW equation (Grassberger et al., 2016, Garcia-Millan et al., 2017, Lallemant et al., 8 Aug 2025).