The Directed Abelian Sandpile Model on Cylinders

Abstract: We study the abelian sandpile model in two dimensions on a directed cylindrical lattice with periodic transverse boundary conditions in the transverse direction and dissipation at one boundary. Recurrent configurations form a finite abelian group, and repeated grain addition at a specific site generates deterministic dynamics on this group. Using Dhar's formulation, the sandpile group is identified with the co-kernel of the reduced directed Laplacian. We show that the group structure admits an exact reduction to a transverse problem, allowing complete determination of its cyclic decomposition. Our results establish a direct connection between the algebraic structure of the sandpile group and the periodicity of the driven dynamics, establishing the manner in which the underlying algebraic structure governs both deterministic and stochastic evolution in directed sandpile.

• The paper reduces the cylindrical sandpile group to the transverse circulant block, proving that its state-space size is D_L^n and grows exponentially as r_+^{nL}.
• The paper identifies parity-dependent cyclic decompositions: odd widths yield two equal Lucas-number factors, while even widths generally involve asymmetric Fibonacci-based factors and require extra Smith-form analysis in exceptional cases.
• The paper connects the group structure to dynamics, showing fixed-drive periods are group-element orders and random-drive return times equal the group size when boundary generators span the group, with simulations confirming exponential growth at exponent 0.96341 ± 0.00061.

Model and geometry

This paper analyzes the directed abelian sandpile model (DASM) on an $n \times L$ cylindrical lattice, periodic in the transverse direction $y$ and dissipative at the downstream boundary $x = n+1$. Each non-sink vertex has out-degree three: toppling sends one grain to $(x+1,y)$ and one grain each to $(x, y\pm 1)$ modulo $L$, with threshold height three. Driving is either fixed (repeated addition at a single boundary site) or random (addition at a uniformly chosen boundary site). The reduced Laplacian $\Delta_{n,L}$ is an $nL \times nL$ block upper-triangular matrix whose diagonal blocks are the circulant matrix $B_L = 3I - A_{C_L}$, where $A_{C_L}$ is the adjacency matrix of the cycle graph on $y$0 vertices, and whose superdiagonal blocks are $y$1. This block structure cleanly separates transverse interactions from directed longitudinal transport.

The sandpile group as a co-kernel

Following Dhar's formulation, recurrent configurations form a finite abelian group identified with the co-kernel of the reduced Laplacian,

$y$2

with cardinality $y$3, where $y$4. Because $y$5 is triangular in blocks, its determinant factors as $y$6, so the longitudinal direction contributes only through the power $y$7 while the transverse operator $y$8 fully determines the group structure. This permits an exact reduction of the $y$9-dimensional problem to the single $x = n+1$0-dimensional transverse block, with the Smith normal form of $x = n+1$1 yielding the complete cyclic decomposition.

The determinant admits a closed form. Since $x = n+1$2 is circulant with eigenvalues $x = n+1$3, one obtains

$x = n+1$4

so that $x = n+1$5 grows exponentially, and $x = n+1$6 satisfies the recurrence $x = n+1$7 with $x = n+1$8, $x = n+1$9. Consequently $(x+1,y)$0.

Parity-dependent cyclic decomposition

A central result is a strong parity dependence of the invariant factors, expressed through Fibonacci numbers $(x+1,y)$1 and Lucas numbers $(x+1,y)$2:

Width	Determinant	Group structure
Odd $(x+1,y)$3	$(x+1,y)$4	$(x+1,y)$5
Even $(x+1,y)$6, $(x+1,y)$7	$(x+1,y)$8	$(x+1,y)$9

For odd widths the group always decomposes into two identical cyclic components — e.g., $(x, y\pm 1)$0 and $(x, y\pm 1)$1. For even widths the decomposition is generally asymmetric; when $(x, y\pm 1)$2 additional cyclic factors appear and the full answer requires the Smith normal form of $(x, y\pm 1)$3 directly, so the even-width classification is conditional rather than complete. The paper also verifies the framework against the known ladder case ($(x, y\pm 1)$4, threshold two), recovering $(x, y\pm 1)$5, consistent with the earlier mapping of ladder dynamics to a random walk on a ring of size $(x, y\pm 1)$6 [Yadav et al., PRE 85, 061114 (2012)].

Dynamical consequences

Under fixed driving at site $(x, y\pm 1)$7, the addition operator acts on recurrent configurations as translation by the group element $(x, y\pm 1)$8, so the orbit period equals $(x, y\pm 1)$9 and is independent of the initial configuration. In the Smith basis, the period is $L$0 over invariant factors $L$1 and generator components $L$2; multiple invariant factors therefore generate multiple characteristic time scales absent in the single-cyclic-factor ladder case.

Under random driving, the evolution is a random walk on $L$3 generated by $L$4. When these generators span the full group, the Markov chain is irreducible with uniform stationary distribution, and Kac's recurrence theorem implies the mean return time to any configuration equals $L$5. Numerical simulations for $L$6 with $L$7 driving steps per width yield a fitted growth exponent $L$8, in close agreement with the exact value $L$9; simulations for $\Delta_{n,L}$0 confirm exponential growth as well. The multi-scale structure induced by the invariant factors provides an algebraic mechanism underlying long-range temporal correlations and $\Delta_{n,L}$1 noise previously observed in directed sandpiles.

Limitations and open questions

Several qualifications apply. The even-width cyclic decomposition is complete only under the coprimality condition $\Delta_{n,L}$2; otherwise additional factors may arise and must be computed case by case. The dynamical analysis establishes periods and return times but does not compute the mixing time of the random walk or the explicit spectral density governing correlation functions, leaving quantitative links between invariant factors and $\Delta_{n,L}$3 noise informal. The reduction relies on the specific block-circulant structure of the cylinder; extension to general directed graphs or non-uniform transverse operators remains open, as does a systematic treatment of how the generators $\Delta_{n,L}$4 fail to span the full group in exceptional cases.

Conclusion

The paper gives an exact algebraic characterization of the sandpile group for the DASM on cylinders of arbitrary circumference, reducing the problem to the transverse circulant block and obtaining closed-form determinants and parity-dependent cyclic decompositions involving Fibonacci and Lucas numbers. It thereby extends Dhar and Ramaswamy's co-kernel framework to a nontrivial quasi-one-dimensional geometry and demonstrates concretely how the same group structure governs both deterministic periodic orbits under fixed driving and ergodic stochastic evolution under random driving, with numerically verified exponential scaling of the state space.