---
title: Non-Abelian Colored Sandpiles
url: https://www.emergentmind.com/topics/non-abelian-colored-sandpiles
type: topic
---

# Non-Abelian Colored Sandpiles

Searching arXiv for the specified topic and closely related papers.
to=arxiv_search.search  ҭыԥjson
{"query":"\"non-Abelian\" colored sandpile sandpile directed trees Colored Sandpile", "max_results": 10, "sort_by": "submittedDate", "sort_order": "descending"}
Inspecting the most relevant returned papers.
to=arxiv_search.search  北京赛车微信json
{"query":"1004.4861 OR 1305.1697 OR 2508.10403", "max_results": 10, "sort_by": "relevance", "sort_order": "descending"}
Non-Abelian colored sandpiles are sandpile models in which grains carry distinct colors or species and the toppling dynamics depend on color, stack order, or local configuration in a way that makes the toppling operators non-commuting. In the explicitly colored lattice model, each color is assigned to a lattice direction, grains move only along their assigned axes, and toppling removes the bottom-most grains in FIFO order, so the relaxed state depends on the sequence of topplings [2508.10403]. In the broader non-Abelian directed-sandpile framework, the defining feature is that toppling depends on the instantaneous local height or on a last-grain bias, which generates spatially correlated metastable patterns, algebraic decay of grain density, and crossover behavior controlled by stochasticity and threshold parity [1004.4861]. On directed trees, nonabelian sandpile dynamics admit an exact nonequilibrium treatment in terms of product-form stationary measures, exact spectra, and monoid methods [1305.1697]. Taken together, these works define the contemporary research landscape for non-Abelian colored sandpiles as a class of order-sensitive, driven-dissipative models with state-dependent transport, nontrivial steady states, and universality behavior distinct from classical Abelian sandpiles.

## 1. Definition and non-Abelian structure

The operator criterion for Abelian symmetry is standard: if a toppling operation at site \(i\) is represented by an operator \(T_i\), then Abelian sandpiles satisfy \([T_i,T_j]=0\), so the final state of a relaxation is independent of toppling order. In non-Abelian sandpiles, toppling depends on the instantaneous local configuration or on the sequence of previous topplings, so \([T_i,T_j]\neq 0\) and the order of topplings matters [1004.4861].

In the colored sandpile model on a hypercubic lattice, the coordination number \(z\) equals the number of colors \(c\), and each color \(a\in\{1,\dots,c\}\) is assigned a unique lattice axis direction. For the square lattice, colors \(1,2,3,4\) correspond to \(+x,+y,-x,-y\), respectively. At each site \(i\), the total height is \(h_i=\sum_{a=1}^c h_i^{(a)}\), and the site carries an ordered stack \(S_i\) of grains in bottom-to-top order. New grains are added on the top, while toppling removes the bottom-most \(n_c\) grains in FIFO order; when site \(i\) topples, the bottom-most \(n_c\) grains are removed and transferred to neighbors according to their colors [2508.10403].

The formal toppling operator is state dependent. If \(m_i^{(a)}\) is the number of grains of color \(a\) among the bottom \(n_c\) elements of \(S_i\), then the per-color update is
\[
h_i^{(a)} \mapsto h_i^{(a)} - m_i^{(a)},\quad
h_{i+v_a}^{(a)} \mapsto h_{i+v_a}^{(a)} + m_i^{(a)}
\quad \text{for all } a.
\]
Unlike the Abelian sandpile model, \(m_i^{(a)}\) depends on the instantaneous stacked order \(S_i\), so the effective toppling matrix is state dependent and the dynamics are non-Abelian [2508.10403].

The one-dimensional counterexample given for \(d=1\), \(c=2\), threshold \(n_c=2\), with initial stacks \(S_k=[1]\), \(S_i=[1,3]\), and \(S_j=[3,1]\), makes the non-commutativity explicit: the sequence \(T_i\), then \(T_j\), then \(T_k\) yields a different emitted color content from \(k\) than the sequence \(T_j\), then \(T_i\), then \(T_k\), and therefore a different fully relaxed configuration and different outflows at the boundaries. The resulting conclusion is \([T_i,T_j]\neq 0\), and the source of the difference is precisely the FIFO rule and the stacking order at the common neighbor [2508.10403].

A broader formulation, used in directed sandpile models, defines non-Abelianity by replacing the Abelian rule \(\Delta_{i,t}=z_c\) with the non-Abelian rule \(\Delta_{i,t}=z(i,t)\), so that the number of toppled grains equals the actual height at toppling. This breaks Abelian symmetry because the local height at the instant of toppling depends on the previous relaxation history [1004.4861]. A plausible implication is that explicitly colored FIFO rules and height-dependent directed rules instantiate the same general mechanism: order-sensitive toppling operators.

## 2. Lattice realizations, driving, and conservation laws

The explicitly colored model is defined on a hypercubic lattice of linear size \(L\) in \(d\) dimensions, with most results shown for the \(d=2\) square lattice and additional results for \(d=1\). The threshold \(n_c\) is fixed; in most two-dimensional simulations \(n_c=4\), while one-dimensional simulations used \(n_c=2\). A site is unstable if \(h_i\ge n_c\), and exactly \(n_c\) grains leave the site when it topples [2508.10403].

The driving protocol is slow addition at uniformly random sites. Each added grain has a color chosen uniformly at random from \(\{1,\dots,c\}\), and the grain is appended to the top of the local stack \(S_i\). Open boundaries are used; when a grain moves off the lattice in its assigned direction at a boundary site, it exits the system and is dissipated [2508.10403]. Reflection symmetry exists about the \(x\) and \(y\) axes with relabeling \(1\leftrightarrow 3\) and \(2\leftrightarrow 4\). If color probabilities are unequal at input, avalanches become anisotropic; equal probabilities produce no overall preferred direction [2508.10403].

The model has exact local bulk conservation. Each toppling removes exactly \(n_c\) grains and transfers exactly those \(n_c\) grains to neighbors; there is no creation or annihilation in the bulk. Colors are also conserved locally: a grain’s color does not change, and per-color counts are advected along their assigned axes. Global grain number and color counts decrease only through boundary outflow [2508.10403].

Directed non-Abelian sandpile models provide a complementary geometry. There, the lattice is a two-dimensional \(\pi/4\)-rotated lattice of sizes \((L_\perp,L_\parallel)\) denoted \((L,T)\), with propagation along layers \(t=0,1,\dots,T-1\), periodic boundary conditions in the transverse direction, and an open boundary at the bottom layer \(t=T-1\). Grains are added at random on the top layer \(t=0\), avalanches propagate from layer \(t\) to \(t+1\), and dissipation occurs only at the bottom [1004.4861]. This directed setting is not a colored model in the strict sense, but the detailed synthesis states that color-dependent rules can be mapped to its non-Abelian mechanisms by taking \(z(i,t)=(z_1,\dots,z_C)\) and allowing color-resolved toppling operators whose last-grain routing depends on internal state, parity, or recent color sequence [1004.4861].

On directed trees, the geometry is an arborescence \(T=(V,E,T)\) with a root \(r\), leaf reservoirs, vertex capacities \(T_v\), and irreversible flow along the unique directed path from any leaf to the root. Grains enter only through leaves, move toward the root, and exit at the root. The state space is \(\Omega(T)=\{t=(t_v)_{v\in V}:0\le t_v\le T_v\}\) [1305.1697]. Here again, the paper does not directly study colored grains, but it gives a rigorous nonequilibrium setting in which nonabelianity can be analyzed exactly.

## 3. Steady states and spatial organization

The colored sandpile has a stationary state with a nontrivial spatial structure. In two dimensions with \(n_c=4\), the average avalanche size satisfies
\[
\langle s(L)\rangle = \frac{L+1}{8},
\]
because a particle dropped at random needs \((L+1)/2\) topplings on average to exit, and each toppling creates four particle jumps. This relation was numerically verified with high precision for \(L\) from \(4\) to \(4096\): \([8\langle s(L)\rangle]/[L+1]=1\pm 2\times 10^{-5}\) [2508.10403].

The total density \(\rho(x,y)\) is not uniform in the stationary state: it peaks at the center and decreases toward boundaries, with larger values at the midpoints of edges than at corners. The average density obeys the finite-size scaling form
\[
\rho(L)=\rho_c + A L^{-1/\nu}.
\]
The reported estimates in two dimensions are \( \rho_c=0.5536,\, 1/\nu=0.503 \) for \(n_c=2\); \( \rho_c=1.0999,\, 1/\nu=0.261 \) for \(n_c=3\); \( \rho_c=1.6426,\, 1/\nu=1.767 \) for \(n_c=4\); and \( \rho_c=2.1856,\, 1/\nu=1.036 \) for \(n_c=5\) [2508.10403]. The occupancy fractions \(f_k\), defined as the fraction of sites with \(k\) grains for \(k=0,\dots,n_c-1\), were also extrapolated for \(L\to\infty\), yielding, for example, \(f_0=0.1973\), \(f_1=0.2460\), \(f_2=0.2729\), and \(f_3=0.2838\) for \(n_c=4\) [2508.10403].

Color-resolved densities exhibit directed gradients. The density of color \(1\) particles, which move along \(+x\), increases roughly linearly with \(x\), while the density of color \(3\) decreases with \(x\). The coarse-grained fluxes satisfy
\[
j_{\kappa}(x+\hat e_{\kappa},y)=j_{\kappa}(x,y)+1/L^2,\qquad
j_1(x,y)=x/L^2,
\]
with analogous expressions for other colors. At site \((x,y)\), outgoing flux components are in the ratio \(x:y:(L-x+1):(L-y+1)\). The synthesis explicitly notes that the naive prediction that average per-color occupancies follow these ratios is approximately, but not exactly, satisfied because of correlations in avalanche color sequences [2508.10403].

The directed non-Abelian framework emphasizes a different stationary signature: the metastable pattern. There the layer-resolved height distribution \(Q(z,t)\), grain density \(\rho(t)\), and occupation density \(\rho_\theta(t)\) are central observables. In non-Abelian directed sandpiles, both \(\rho(t)\) and \(\rho_\theta(t)\) decay algebraically along the propagation direction,
\[
\rho(t)\sim \rho_\theta(t)\sim t^{-\alpha},
\]
whereas in Abelian directed sandpiles the metastable patterns are uncorrelated and \(\rho(t)\) is flat, corresponding to \(\alpha=0\) [1004.4861]. This suggests a sharp distinction between Abelian and non-Abelian colored systems: nonuniform or algebraically structured stationary profiles are not incidental finite-size effects but operational diagnostics of order-sensitive dynamics.

## 4. Avalanche statistics and universality classes

For the colored sandpile in two dimensions, the avalanche size distribution obeys the finite-size scaling form
\[
D(s,L)L^\beta \sim \mathcal{G}(s/L^\alpha),
\]
with \(\beta=3\), \(\alpha=2\), and therefore \(\tau=\beta/\alpha=1.5\). Data collapse was reported for \(L=1024\), \(4096\), and \(16384\) by plotting \(D(s,L)L^3\) versus \(s/L^2\) [2508.10403]. The exponents are described as nearly independent of where the avalanche is triggered, including corners, sides, and center [2508.10403].

The duration distribution uses the avalanche life-time \(T\), defined as the number of synchronous update steps until stability when all currently unstable sites topple in randomly permuted order and all emitted particles jump simultaneously. The reported scaling collapse is obtained by plotting \(D(T,L)L^{1.84}\) versus \(T/L\), implying
\[
D(T)\sim T^{-\tau_T}
\quad\text{with}\quad
\tau_T\approx 1.84,
\]
and the average duration scales as
\[
\langle T(L)\rangle \sim L^{0.1474}.
\]
In one dimension, the size distribution has \(\beta=2.0\), \(\alpha=1.5\), and therefore \(\tau=4/3\), while the duration distribution has \(\beta_T=1.5\), \(\alpha_T=1\), and therefore \(\tau_T=1.5\) [2508.10403].

The paper explicitly states that the colored sandpile belongs to a different universality class of sandpile models. The reason given is the combination of directed per-grain motion, state-dependent toppling content \(m_i^{(a)}\) due to FIFO stacks, and resulting non-Abelian dynamics, which produce critical exponents and spatial structures distinct from Abelian BTW and commonly studied stochastic sandpiles such as Manna [2508.10403].

The directed non-Abelian sandpile framework supplies a complementary universality classification. For avalanche observables \(x\in\{s,t,a,w,h\}\), the distributions satisfy
\[
P(x;T)\sim x^{-\tau_x} f(x/T^{D_x}),
\]
with \(D_t=1\), \(\langle s\rangle\sim T\), \(D_s(2-\tau_s)=1\), \(D_a=D_w+1\), and \(D_s=D_a+D_h\) [1004.4861]. In the non-Abelian stochastic class, the metastable decay exponent is numerically \(\alpha\approx 0.45\approx 1/2\) for \(z_c=2\), and the avalanche exponents match those of Abelian stochastic sandpiles: \(\tau_s=10/7\), \(\tau_t=7/4\), and \(D_h=1/4\) [1004.4861]. In the mean-field-like non-Abelian deterministic class, \(\alpha\to 1\), \(\tau_s=3/2\), and \(\tau_t=2\) [1004.4861].

A precise caution follows from these results: avalanche exponents alone can be ambiguous. The same synthesis states that mass and duration exponents can be mean-field-like even when metastable correlations differ, and that \(\alpha\) and the broadness of \(Q_r(z,t)\) are needed to resolve the universality class [1004.4861]. This is directly relevant to colored models because the explicitly colored paper reports exponents and spatial structure, while the directed non-Abelian paper identifies metastable diagnostics that distinguish stochastic-like from mean-field-like order-sensitive dynamics.

## 5. Metastable patterns, threshold parity, and crossover behavior

The central conceptual contribution of the directed non-Abelian work is that non-Abelianity induces spatially correlated metastable patterns characterized by the algebraic decay of grain density along the propagation direction. The scar density \(\rho_{sc}(t)\), which counts recent avalanche boundary traces, obeys \(\rho_{sc}(t)\sim w(t)^{-1}\), so \(\alpha_{sc}=D_w\). In non-Abelian directed sandpiles, grains remain only along boundaries, implying \(\alpha=\alpha_{sc}\); in Abelian directed sandpiles, \(\alpha=0\) even though \(\alpha_{sc}=1/2\) [1004.4861].

Threshold parity is decisive. For deterministic non-Abelian directed models with even \(z_c\), last-grain odd-event bias is diluted by the many even splits, and the system tends toward mean-field-like behavior. For odd \(z_c\), the last-grain choice becomes relevant and can induce random-walk-like avalanche boundaries and stochastic universality behavior, even with deterministic rules, especially in the alternative-bias and partial-bias rules near \(p=1/2\) [1004.4861].

The paper gives explicit crossover forms. For the alternative-bias rule with odd \(z_c\),
\[
\rho_\theta(t)=t^{-1} g(t/t_\times),
\]
with \(g(x)\sim O(1)\) for \(x\ll 1\), \(g(x)\sim x^{1/2}\) for \(x\gg 1\), and
\[
t_\times \sim z_c^2.
\]
For the partial-bias rule at fixed odd \(z_c\), for example \(z_c=3\), and bias \(p\) close to \(1\), one finds
\[
\rho_\theta(t)\sim (1-p)^{1/2}(t(1-p)^{1/2})^{-1} g(t(1-p)^{1/2}),
\]
so
\[
t_\times \sim (1-p)^{-1/2}.
\]
In the non-Abelian stochastic model, increasing \(z_c\) suppresses effective stochasticity and yields a crossover from \(\alpha=1/2\) to \(\alpha=1\) at fixed \(t\), while at fixed \(z_c\), \(\rho_\theta(t)\) can cross from \(\alpha=1\) to \(\alpha=1/2\) with \(t\) [1004.4861].

To diagnose these regimes, the paper defines the rescaled distribution
\[
Q_r(z,t)=\frac{Q(z,t)}{1-Q(0,t)},
\]
supported on \(z\ge \lfloor z_c/2\rfloor\). Broad \(Q_r\) indicates stochastic universality, while sharply peaked \(Q_r\) around \(\lfloor z_c/2\rfloor\) indicates mean-field-like deterministic behavior with ballistic boundaries [1004.4861].

The detailed synthesis makes the extension to colored sandpiles explicit but carefully qualified. It states that the metastable-pattern signatures and avalanche universality classes carry over to colored variants when color-resolved routing is stochastic or when deterministic color routing has a last-grain bias. Stochastic color routing is associated with non-Abelian stochastic universality, \(\alpha\approx 1/2\), \(\tau_s=10/7\), and \(\tau_t=7/4\), whereas deterministic color routing with even effective threshold is associated with mean-field-like deterministic behavior, \(\alpha\to 1\), \(\tau_s=3/2\), and \(\tau_t=2\) [1004.4861]. This is presented as a mapping rather than a direct result for explicit colored grains. A plausible implication is that parity-like effects in colored models arise whenever deterministic color-specific splits leave one “last” grain whose routing controls whether boundaries are ballistic or diffusive.

## 6. Trees, monoids, and exact nonequilibrium results

Directed nonabelian sandpile models on trees provide a mathematically exact setting for noncommutative sandpile dynamics. Two classes are defined: the Trickle-down sandpile model, in which sand grains are allowed to move one at a time, and the Landslide sandpile model, in which all the grains at a vertex topple at once [1305.1697]. In both cases, grains enter only through specified reservoirs at leaves, move along the unique path to the root, and the order of topplings matters because a moved grain may fill a site needed by a subsequent move; in general \(\theta_u\theta_v\neq \theta_v\theta_u\) and \(\tau_u\tau_v\neq \tau_v\tau_u\) [1305.1697].

For the Trickle-down model, the Markov chain is ergodic and the stationary distribution is of product form. If
\[
L_v := \{\ell\in L : v\in \ell^\downarrow\},\qquad
Y_v := \sum_{\ell\in L_v} y_\ell,
\]
then the one-site marginal is
\[
\rho_v(h):=\frac{Y_v^h\, x_v^{T_v-h}}{\sum_{i=0}^{T_v} Y_v^i\, x_v^{T_v-i}},
\qquad 0\le h\le T_v,
\]
and the full stationary distribution factorizes as
\[
P(t)=\prod_{v\in V}\rho_v(t_v).
\]
The nonequilibrium partition function is
\[
Z_\theta=\prod_{v\in V}\left(\sum_{i=0}^{T_v} Y_v^i\, x_v^{T_v-i}\right)
\]
[1305.1697].

For the Landslide model, the stationary distribution is not a product measure in general, but the spectrum is exactly solvable. Writing
\[
x_S := \sum_{v\in S} x_v,\qquad
y_S := \sum_{\substack{\ell\in L\\ \ell^\downarrow\subseteq S}} y_\ell,\qquad
T_S := \prod_{v\in S} T_v,
\]
the characteristic polynomial of the transition matrix is
\[
\det(M_\tau-\lambda I)=\prod_{S\subseteq V} (\lambda-y_S-x_S)^{\,T_{S^c}},
\qquad S^c=V\setminus S.
\]
Thus every eigenvalue is a partial sum \(\lambda_S=x_S+y_S\), with multiplicity \(T_{S^c}\), and the zero eigenvalue \(\lambda_\varnothing=0\) appears with multiplicity \(T_V\) [1305.1697].

The method is algebraic. The trickle-down monoid \(N(T)=\langle {}_v,\theta_v:v\in V\rangle\) and the landslide monoid \(M(T)=\langle {}_v,\tau_v:v\in V\rangle\) admit recursive decompositions under removal of a leaf, and the landslide monoid is \(R\)-trivial. The proofs use wreath products and the representation theory of monoids [1305.1697]. The convergence rate is also explicit: with \(p_x:=\min_{v\in V} x_v\) and \(n:=|V|\), for \(k\ge (n-1)/p_x\),
\[
\|P^k-\pi\| \le \exp\!\left(- \frac{(k p_x - (n-1))^2}{2 k p_x}\right),
\]
and a mixing-time bound to distance at most \(e^{-c}\) is
\[
\text{mixing time} \le \frac{2(n+c-1)}{p_x}.
\]
These models shift emphasis from avalanche criticality to exact stationary measures and spectral analysis of driven, noncommutative Markov chains [1305.1697].

The same synthesis proposes a natural colored extension, but it is explicitly identified as a proposed extension rather than a direct theorem. If toppling is color-blind and capacities are shared across colors, then the total-occupancy projection coincides with the uncolored model, the total-occupancy eigenvalues remain
\[
\lambda_S^{\text{tot}} = x_S + \sum_{\substack{\ell\in L\\ \ell^\downarrow\subseteq S}} \sum_{c=1}^C y_\ell^{(c)},
\]
with multiplicities \(T_{S^c}\), and the trickle-down stationary distribution plausibly retains vertex-wise independence for total occupancy with multinomial color composition conditional on height [1305.1697]. This suggests a rigorous route to colored non-Abelian sandpiles in nonequilibrium settings when the color dynamics preserve \(R\)-triviality.

## 7. Comparisons, misconceptions, and scope

A recurrent misconception is that adding colors merely decorates an Abelian sandpile. The explicitly colored model contradicts this directly: because toppling removes the bottom-most grains in FIFO order, the emitted color content depends on the toppling sequence, the effective toppling matrix is state dependent, and the final stable state depends on toppling order [2508.10403]. Likewise, in directed and tree-based non-Abelian sandpiles, topplings do not commute because the local height or the first available capacity along a path depends on previous moves [1004.4861; 1305.1697].

A second misconception is that directional micro-motion necessarily implies a globally directed avalanche. The colored sandpile assigns one axis to each color and therefore breaks rotational symmetry microscopically, but with equal color probabilities the avalanches or, more generally, the self-organization processes have no overall preferred direction, and avalanche propagation is equally effective in all directions [2508.10403]. This distinguishes the model from exactly solved directed sandpiles, which have an overall preferred direction [2508.10403].

A third misconception is that the BTW limit is recovered by collapsing to one color. The colored-sandpile synthesis states explicitly that BTW is not recovered by taking all grains the same color in this model, because directed motion along a single axis and FIFO sequence effects remain [2508.10403]. More generally, the detailed comparison notes that Abelian sandpile models have a fixed, linear, state-independent toppling matrix and compact nested multiply-toppled regions, whereas colored non-Abelian avalanches are non-compact and fragmented [2508.10403].

The relation among the three research strands is therefore complementary rather than redundant. The explicitly colored lattice model establishes that distinguishable grains moving along fixed axes and interacting through non-Abelian toppling produce new steady states, nontrivial spatial structure, and exponents such as \(\tau=1.5\) in two dimensions [2508.10403]. The directed non-Abelian framework identifies the metastable observables, crossover laws, and parity effects that organize non-Abelian universality classes [1004.4861]. The tree models show that noncommutative sandpile dynamics can also be treated exactly through product-form stationary measures, exact spectra, and \(R\)-trivial monoids [1305.1697]. Taken together, these results support a precise use of the term “non-Abelian colored sandpiles”: sandpile systems with color-resolved transport and order-sensitive relaxation in which metastable structure, not only avalanche exponents, is central to classification.

Source: https://www.emergentmind.com/topics/non-abelian-colored-sandpiles