Algebraic Limits of Sandpiles

Mikhail Shkolnikov · 2025

Read Algebraic Limits of Sandpiles by Mikhail Shkolnikov (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-09912.txt.

Introduction

Mikhail Shkolnikov builds algebra for sandpiles, the toy model behind self-organized criticality, the idea that some systems tune themselves to a critical edge. In the model, grains pile on a grid until slopes topple, and the stable patterns form a group with rich structure. A long-open question asked what that group becomes in the limit, as the grid grows without bound.

The paper answers the question for the extended sandpile group and ties it to a second limit, taken across the classical sandpile groups with maps that embed one in the next. Shkolnikov also proposes canonical maps running the other way, called epimorphisms. An unexpected payoff follows for pure number theory: new divisibility properties for the counts of spanning trees on rectangular grids.

Catalog record

CREATORMikhail Shkolnikov
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-09912

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Algebraic Limits of Sandpiles Mikhail Shkolnikov Abstract The paper contributes to building algebraic foundations of self-organized criticality answering a previously unsolved question about the limiting structure of the extended sandpile group as well as relating it to another limit at the level of classical sandpile groups with respect to certain monomorphisms, and puts forward a concept of canonical sandpile epimorphisms, drawing an unexpected consequence about the divisibility properties of the numbers of spanning trees on rectangles. 1 Introduction What we now call “sandpile model” has appeared multiple times in diverse guises throughout mathematics and physics, but has acquired a tremendous attention of a large scientific community only when it was proposed as a prototype for self-organized criticality [ 1 ] . The name of the model used in the present note was coined by Deepak Dhar with an additional adjective “Abelian” referring in part to his discovery [ 2 ] of the fact that the set of recurrent states of the model on a finite domain has a natural structure of an Abelian group. One characteristic feature of self-organized critical systems is their scale transcendence, which manifests in the form of various scaling limit theorems such as [ 3 ] or [ 4 ] . These type of results may be described as geometric or “visual” approach to the limiting structure of sandpiles, since the graphical representation of states is under consideration. Alternatively, one may focus on probabilistic properties of the sandpile model in the scaling limit (see for instance [ 5 ] ) ignoring altogether how the states look like. What is going to be discussed next is similar to the latter spirit, but the attention will be turned to the algebraic aspects of the theory, i.e. our goal is to prepare the ground for the abstract understanding of sandpile groups in the limit, ignoring for the moment the intricate structure and stunning beauty of the underlying recurrent configurations. Perhaps, over time, these three approaches – geometric, probabilistic and algebraic – will be unified. 2 Sandpile groups and harmonic functions Consider a finite connected undirected graph without loops with a distinguished vertex that we call the sink. Denote by \Gamma the set of non-sink vertices of this graph. A state of the sandpile model is a non-negative integer-valued function on \Gamma. One may think that a value of a state at a vertex represents the number of sand grains. The (reduced) Laplacian \Delta\colon\mathbb{Z}^{\Gamma}\rightarrow\mathbb{Z}^{\Gamma} is given by the difference of the symmetric adjacency matrix and the diagonal degree matrix with removed row and column corresponding to the sink. Note that the graph is restored from its Laplacian, the absolute value of its determinant is equal to the number of spanning trees on the graph by Kirchhoff’s theorem. A state \phi is called stable if \phi(v) is less than the degree of v for all v\in\Gamma. If a state is unstable there exist a vertex v where \phi(v) is greater or equal to the degree of v and a toppling (i.e. redistribution of sand grains) \phi\mapsto\phi+\Delta\delta_{v} can be performed. Here \delta_{v} is a function on \Gamma equal to 1 at v and vanishing otherwise. A relaxation is a process of applying topplings until the state becomes stable. The process is always finite and the resulting state \phi^{\circ} doesn’t depend on the order of topplings. A stable state is called recurrent if it can be obtained from any other state by adding sand grains and relaxing. The sandpile group G(\Gamma) consists of recurrent states with an operation given by (\phi,\psi)\mapsto(\phi+\psi)^{\circ}. For a function f\in\mathbb{Z}^{\Gamma} there exist a unique recurrent state \phi and a function H\in\mathbb{Z}^{\Gamma} such that f=\phi+\Delta H. This gives an isomorphism G(\Gamma)\cong\operatorname{Coker}(\Delta)=\mathbb{Z}^{\Gamma}/\penalty\Delta(\mathbb{Z}^{\Gamma}) which can be written equivalently in a form of a short exact sequence 0\rightarrow\mathbb{Z}^{\Gamma}\xrightarrow{\Delta}\mathbb{Z}^{\Gamma}\rightarrow G(\Gamma)\rightarrow 0, where the notaion means that the left homomorphism is injective (which follows from the non-vanishing of the determinant of \Delta ), the right one is surjective, and the image and kernel coincide in the middle. Denote by \partial\Gamma\subset\Gamma the set of vertices adjacent to the sink. We call \partial\Gamma the boundary and \Gamma^{\circ}=\Gamma\backslash\partial\Gamma the interior of the graph. The extended sandpile group \widetilde{G}(\Gamma) is derived by allowing the recurrent states to take real non-negative values along \partial\Gamma . In other words, there is a short exact sequence 0\rightarrow\mathbb{Z}^{\Gamma}\xrightarrow{\widetilde{\Delta}}\mathbb{Z}^{\Gamma^{\circ}}\mkern-9.0mu\oplus\mathbb{R}^{\partial\Gamma}\rightarrow\widetilde{G}(\Gamma)\rightarrow 0, where \widetilde{\Delta} is the composition of \Delta with the inclusion of \mathbb{Z}^{\Gamma}=\mathbb{Z}^{\Gamma^{\circ}}\mkern-9.0mu\oplus\mathbb{Z}^{\partial\Gamma} to \mathbb{Z}^{\Gamma^{\circ}}\mkern-9.0mu\oplus\mathbb{R}^{\partial\Gamma}. There is another short exact sequence 0\rightarrow G(\Gamma)\rightarrow\widetilde{G}(\Gamma)\xrightarrow{/\penalty\mathbb{Z}}(\mathbb{R}/\penalty\mathbb{Z})^{\partial\Gamma}\rightarrow 0, where /\penalty\mathbb{Z} takes the fractional part of the extended recurrent state and \mathbb{R}/\penalty\mathbb{Z} is the quotient of \mathbb{R} by the additive subgroup \mathbb{Z} which topologically is a circle. Let \Delta^{\circ}\colon\mathbb{Z}^{\Gamma}\rightarrow\mathbb{Z}^{\Gamma^{\circ}} be a map given by the composition of \Delta with the restriction to \Gamma^{\circ}. For an Abelian group A define the group \mathcal{H}_{A}(\Gamma) of A -valued harmonic functions as the kernel of \Delta^{\circ}\otimes A. Consider a category \mathcal{D}\mkern-12.0mu\mathcal{D} of connected graphs without loops (possibly infinite, with or without sink) represented by pairs (\Gamma,\Delta) with morphisms given by embeddings \iota\colon\Gamma_{1}\rightarrow\Gamma_{2} such that \Delta_{1} is obtained from \Delta_{2} by removing rows and columns corresponding to vertices in \Gamma_{2}\backslash\iota(\Gamma_{1}). For example, the casual map \Omega\subset\mathbb{R}^{n}\mapsto\Omega\cap\mathbb{Z}^{n} is upgraded to a contravariant functor from the category of open sets in \mathbb{R}^{n} with inclusions as morphisms to \mathcal{D}\mkern-12.0mu\mathcal{D}. Note that A^{\Gamma} and A^{\Gamma^{\circ}} are contravariant functors from \mathcal{D}\mkern-12.0mu\mathcal{D} to Abelian groups and \Delta^{\circ}\otimes A is their natural transformation. Therefore, \mathcal{H}_{A}(\Gamma) is also a contravariant functor, with the induced homomorphisms being simple restrictions. Proposition 1 . The groups \widetilde{G}(\Gamma) and \mathcal{H}_{\mathbb{R}/\penalty\mathbb{Z}}(\Gamma) are canonically isomorphic. Proof. The isomorphism is given by the connecting homomorphism arising in the following employment of the snake lemma where for the middle row we use the invertibility of \Delta\otimes\mathbb{R}: {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces\mathbb{Z}^{\Gamma}} {\lx@inpgf@ignorespaces\mathbb{Z}^{\Gamma^{\circ}}\mkern-9.0mu\oplus\mathbb{R}^{\partial\Gamma}} {\lx@inpgf@ignorespaces\widetilde{G}(\Gamma)} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces\mathbb{R}^{\Gamma}} {\lx@inpgf@ignorespaces\mathbb{R}^{\Gamma}} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces\mathcal{H}_{\mathbb{R}/\penalty\mathbb{Z}}(\Gamma)} {\lx@inpgf@ignorespaces(\mathbb{R}/\penalty\mathbb{Z})^{\Gamma}} {\lx@inpgf@ignorespaces(\mathbb{R}/\penalty\mathbb{Z})^{\Gamma^{\circ}}} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces 0} \scriptstyle\widetilde{\Delta} \scriptstyle\Delta\otimes\mathbb{R} \scriptstyle\Delta^{\circ}\otimes\mathbb{R}/\penalty\mathbb{Z} ∎ Remark 1 . In a similar way, one obtains G(\Gamma)\cong\operatorname{Ker}(\Delta\otimes\mathbb{R}/\penalty\mathbb{Z}), the latter may be thought of as the group of “strictly harmonic” circle-valued functions on the graph, i.e. the vanishing of the Laplacian is required both in the interior and on the boundary. Another application of the snake lemma gives the following. Proposition 2 . The groups of integer-, real- and circle- valued harmonic functions and the group \operatorname{Coker}\Delta^{\circ} are linked through an exact sequence 0\rightarrow\mathcal{H}_{\mathbb{Z}}(\Gamma)\rightarrow\mathcal{H}_{\mathbb{R}}(\Gamma)\rightarrow\mathcal{H}_{\mathbb{R}/\penalty\mathbb{Z}}(\Gamma)\rightarrow\operatorname{Coker}\Delta^{\circ}\rightarrow 0. Lie groups \mathcal{H}_{\mathbb{R}}(\Gamma) and \mathcal{H}_{\mathbb{R}/\penalty\mathbb{Z}}(\Gamma) both have dimension equal to |\partial\Gamma| and \operatorname{Coker}\Delta^{\circ} is discrete. Therefore, since \mathcal{H}_{\mathbb{R}}(\Gamma) is connected (being an \mathbb{R} -vector space), the image of \mathcal{H}_{\mathbb{R}}(\Gamma) in \mathcal{H}_{\mathbb{R}/\penalty\mathbb{Z}}(\Gamma) is identified with the maximal connected subgroup of \widetilde{G}(\Gamma) that we denote by \widetilde{G}_{0}(\Gamma). We observe the isomorphisms \widetilde{G}_{0}(\Gamma)\cong\mathcal{H}_{\mathbb{R}}(\Gamma)/\penalty\mathcal{H}_{\mathbb{Z}}(\Gamma) and \operatorname{Coker}\Delta^{\circ}\cong\widetilde{G}(\Gamma)/\penalty\widetilde{G}_{0}(\Gamma). Yet another snake lemma diagram (which is also omitted) explains the relation between the cokernel of \Delta^{\circ} and the usual sandpile group. Proposition 3 . There is an exact sequence 0\rightarrow\mathcal{H}_{\mathbb{Z}}(\Gamma)\rightarrow\mathbb{Z}^{\partial\Gamma}\rightarrow G(\Gamma)\rightarrow\operatorname{Coker}\Delta^{\circ}\rightarrow 0. Here, the map \mathcal{H}_{\mathbb{Z}}(\Gamma)\rightarrow\mathbb{Z}^{\partial\Gamma} is the composition of \Delta and the projection to \partial\Gamma. Denote by G_{0}(\Gamma) the image of the homomorphism \mathbb{Z}^{\partial\Gamma}\rightarrow G(\Gamma). It is a subroup of the sandpile group generated by adding sand at the boundary. Note that G_{0}(\Gamma)=\widetilde{G}_{0}(\Gamma)\cap G(\Gamma) and \operatorname{Coker}\Delta^{\circ}\cong G(\Gamma)/\penalty G_{0}(\Gamma). This implies the following lemma. Lemma 1 . If G(\Gamma) is generated by adding sand at the

Read the complete text