Generalized cluster algorithms for Potts lattice gauge theory

Anthony E. Pizzimenti, Paul Duncan, Benjamin Schweinhart · 2025

Read Generalized cluster algorithms for Potts lattice gauge theory by Anthony E. Pizzimenti, Paul Duncan, Benjamin Schweinhart (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-2507-13503.txt.

Introduction

Anthony Pizzimenti, Paul Duncan, and Benjamin Schweinhart extend fast sampling methods to Potts lattice gauge theory. Gauge theories place their variables on the links of a grid, which blocks older cluster tricks. Their 2025 paper recasts the model on flat faces of the grid. That shift lets two classic methods, Swendsen-Wang and invaded-cluster, run in the new setting.

The invaded-cluster version aims itself at the critical point. It stops when a surface first spans the space, a signal read from the shape of clusters. Tests ran on four-dimensional tori for two small gauge groups. Both methods lost memory of past states far faster than single-spin updates. The gain held on grids of scale at least 40. The work opens larger gauge studies to cluster sampling.

Catalog record

CREATORAnthony E. Pizzimenti, Paul Duncan, Benjamin Schweinhart
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-13503

Browse the full library · About the Tower of Babel

Full text

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

Generalized cluster algorithms for Potts lattice gauge theory Anthony E. Pizzimenti, Paul Duncan, Benjamin Schweinhart Abstract Monte Carlo algorithms, like the Swendsen–Wang and invaded-cluster , sample the Ising and Potts models asymptotically faster than single-spin Glauber dynamics do. Here, we generalize both algorithms to sample Potts lattice gauge theory by way of a 2 2 -dimensional cellular representation called the plaquette random-cluster model . The invaded-cluster algorithm targets Potts lattice gauge theory at criticality by implementing a stopping condition defined in terms of homological percolation , the emergence of spanning surfaces on the torus. Simulations for ℤ 2 \mathbb{Z}_{2} and ℤ 3 \mathbb{Z}_{3} lattice gauge theories on the cubical 4 4 -dimensional torus indicate that both generalized algorithms exhibit much faster autocorrelation decay than single-spin dynamics and allow for efficient sampling on 4 4 -dimensional tori of linear scale at least 40 40 . Introduction The q -state Potts lattice gauge theory (PLGT) on a finite subcomplex X of the cubical complex \mathbb{Z}^{d} is a random assignment of spins in \mathbb{Z}_{q} to the edges of X . These assignments are induced by the Hamiltonian \H\left(f\right)=-\sum_{x\in X}\mathbf{1}_{\delta^{1}f\left(x\right)=0}, where x\in X is a plaquette ( 2 -dimensional cell) and \delta^{1}f\left(x\right) is the oriented sum (mod q ) of spins on the edges incident to x [ 33 , 26 ] . Recent work in [ 24 , 12 , 31 ] generalizes the Fortuin–Kastelyn random-cluster model (a graphical representation of the Potts model) to construct a 2 -dimensional cellular representation of Potts lattice gauge theory called the plaquette random-cluster model (PRCM) [ 17 ] . We use this cellular representation to define non-local Monte Carlo algorithms for Potts lattice gauge theory. Specifically, we generalize the Swendsen–Wang [ 32 ] and invaded-cluster algorithms [ 28 , 27 ] , procedures that revolutionized computational simulations of the Potts model. Both algorithms alternate between sampling 2 -subcomplexes and sampling spins: the former does so according to the PRCM and PLGT; 1 1 1 An algorithm of this form was previously described for the special case q=2 in [ 7 , 4 ] . the latter builds a 2 -complex cell-by-cell until a stopping condition based on homological percolation — the emergence of “giant surfaces” — is met, then re-samples spins and begins anew. c d a b Figure 1: A cubical complex with fourteen 0 -cells (vertices), nineteen 1 -cells (edges), and four 2 -cells (filled squares). Edge coefficients are 0 (solid) or 1 (dashed). In Section 2 , we define Potts lattice gauge theory, the plaquette random-cluster model, and the coupling between them. Section 3 covers the Swendsen–Wang and invaded-cluster algorithms, including a proof that the former has the correct stationary distribution. We then discuss consequences of the duality properties of the PRCM in Section 4 ; the Loop–Cluster coupling [ 34 ] and its generalization [ 22 ] are recovered by considering the behavior of the coupling between Potts (hyper)lattice gauge theory and the plaquette random cluster model under the latter’s duality transformation. In Section 5 , we conclude by describing practical implementations of the Swendsen–Wang and invaded-cluster algorithms and analyzing autocorrelation decay in computational simulations. Background and definitions A finite cubical complex X is a combinatorial object made up of vertices, edges, and plaquettes (squares). Typically, X is a finite subset of \mathbb{Z}^{d} , the tiling of \mathbb{R}^{d} by unit d -cubes with corners at integer coordinates. Elements of \mathbb{Z}^{d} are i -cells , where i indicates dimension: 0 -cells are vertices of the integer lattice, 1 -cells are edges in the nearest-neighbors graph on the vertices, and 2 -cells the (square) minimal cycles formed by the edges. In this context, “ 2 -cells,” “plaquettes,” and “ 2 -plaquettes” are the same. We are mostly interested in cells of dimension less than 3 (though 3 -cells enter into the definition of the plaquette invaded-cluster algorithm). To impose periodic boundary conditions on finite complexes, we “glue” opposite cells of [0,N]^{d}\subset\mathbb{Z}^{d} together to form the scale- N d -fold cubical torus \mathbb{T}^{d}_{N} . See Figure 1 for an illustration. Potts lattice gauge theory The q -state Potts lattice gauge theory (PLGT) on a cubical complex X assigns spins in \mathbb{Z}_{q} to the edges of X . 2 2 2 Counter to tradition, it is convenient for us to identify \mathbb{Z}_{q}\coloneqq\left\{0,1,\dots,q-1\right\} with the group of integers modulo q rather than the q th complex roots of unity. As our techniques are inspired by algebraic topology, we assume its vocabulary and refer to the state space of Potts lattice gauge theory as the group of 1 -cochains , denoted by C^{1}(X;\mathbb{Z}_{q}) . Each f\in C^{1}(X;\mathbb{Z}_{q}) is a function that assigns a spin to each edge of X so that the sign of the spin flips whenever its corresponding edge’s orientation is reversed. When q is prime, C^{1}(X;\mathbb{Z}_{q}) is a \mathbb{Z}_{q} -vector space. 3 3 3 Cochains in C^{1}(X;\mathbb{Z}_{q}) are sometimes referred to as discrete differential forms , as they play the same role in cubical cohomology theory as differential forms do in de Rham cohomology. The coboundary (or discrete exterior derivative ) of a 1 -cochain is a 2 -cochain — a map assigning \mathbb{Z}_{q} spins to plaquettes — and belongs to the group C^{2}(X;\mathbb{Z}_{q}) of 2 -cochains on X . For f\in C^{1}(X;\mathbb{Z}_{q}) and a plaquette x , the coboundary \delta^{1}f assigns to x the oriented sum of the spins on its bounding edges. The plaquette x=\{abdc\} in Figure 1 , for example, is oriented so that its boundary is \partial_{2}(x)=\{ab\}-\{ac\}+\{bd\}-\{cd\} and, extending linearly, we define \delta^{1}f(x) by \displaystyle(\delta^{1}f)(x) \displaystyle=f(\{ab\})-f(\{ac\})+f(\{bd\})-f(\{cd\}) \displaystyle=f(\partial_{2}(x)). We can also take the coboundary of a 0 -cochain g\in C^{0}(X;\mathbb{Z}_{q}) , an assignment of spins to the vertices of X . If e is an oriented edge from v to w , then \delta^{0}g\left(e\right)=g\left(w\right)-g\left(v\right) . As \delta^{1}\circ\delta^{0}=0 , the Hamiltonian is unchanged by the addition of the coboundary of a 0 -cochain: this is the action of gauge transformations on C^{1}\left(X;\mathbb{Z}_{q}\right) . See Appendix A.1 for complete definitions. The q -state Potts lattice gauge theory \nu on X with inverse temperature \beta is then the Gibbs distribution on C^{1}\left(X;\mathbb{Z}_{q}\right) induced by the Hamiltonian \H that counts the number of non-frustrated plaquettes x\in X (i.e. plaquettes x such that \delta^{1}f(x)=0 ) [ 33 , 26 ] given by \H\left(f\right)=-\sum_{x\in X}\mathbf{1}_{\delta^{1}f\left(x\right)=0}. In all, PLGT has law \nu(f)=\frac{1}{Z(q,\beta)}e^{-\beta\H(f)}, assigning the probability \nu\left(f\right) to the edge spin assignment f . We can extend this model to define a random i -cochain in C^{i}\left(X;\mathbb{Z}_{q}\right) , where the case i=0 is the classical q -state Potts model [ 33 , 24 ] . Potts lattice gauge theory has been studied as a relatively simple discrete example of a model with a gauge symmetry: when q=2 or q=3 , it coincides with \mathbb{Z}\left(q\right) Euclidean lattice gauge theory (with \mathbb{Z}\left(q\right) the complex q th roots of unity). In addition, Ising lattice gauge theory — the case q=2 — has met renewed interest because of its relationship with the toric code [ 1 ] , spurring on development of fast-converging algorithms for sampling these models. Plaquette random-cluster model To divine the plaquette random-cluster model , we use the same strategy as for the random-cluster model: expand and manipulate the partition function

Read the complete text