Table of Contents
Fetching ...

Homology-changing percolation transitions on finite graphs

Michael Woolls, Leonid Pryadko

Abstract

We consider homological edge percolation on a sequence $(\mathcal{G}_t)_t$ of finite graphs covered by an infinite (quasi)transitive graph $\mathcal{H}$, and weakly convergent to $\mathcal{H}$. Namely, we use the covering maps to classify $1$-cycles on graphs $\mathcal{G}_t$ as homologically trivial or non-trivial, and define several thresholds associated with the rank of thus defined first homology group on the open subgraphs. We identify the growth of the homological distance $d_t$, the smallest size of a non-trivial cycle on $\mathcal{G}_t$, as the main factor determining the location of homology-changing thresholds. In particular, we show that the giant cycle erasure threshold $p_E^0$ (related to the conventional erasure threshold for the corresponding sequence of generalized toric codes) coincides with the edge percolation threshold $p_{\rm c}(\mathcal{H})$ if the ratio $d_t/\ln n_t$ diverges, where $n_t$ is the number of edges of $\mathcal{G}_t$, and we give evidence that $p_E^0<p_{\rm c}(\mathcal{H})$ in several cases where this ratio remains bounded, which is necessarily the case if $\mathcal{H}$ is non-amenable.

Homology-changing percolation transitions on finite graphs

Abstract

We consider homological edge percolation on a sequence of finite graphs covered by an infinite (quasi)transitive graph , and weakly convergent to . Namely, we use the covering maps to classify -cycles on graphs as homologically trivial or non-trivial, and define several thresholds associated with the rank of thus defined first homology group on the open subgraphs. We identify the growth of the homological distance , the smallest size of a non-trivial cycle on , as the main factor determining the location of homology-changing thresholds. In particular, we show that the giant cycle erasure threshold (related to the conventional erasure threshold for the corresponding sequence of generalized toric codes) coincides with the edge percolation threshold if the ratio diverges, where is the number of edges of , and we give evidence that in several cases where this ratio remains bounded, which is necessarily the case if is non-amenable.

Paper Structure

This paper contains 21 sections, 21 theorems, 43 equations, 5 figures, 2 tables.

Key Result

Lemma \oldthetheorem

Consider a finite graph $\mathcal{G}$ covered by an infinite graph $\mathcal{H}$, with the injectivity radius $r_f$. Then the minimum weight $d_Z$ of a non-trivial cycle on $\mathcal{G}$ satisfies the inequality $2r_f+1\le d_Z\le 2r_f+3$.

Figures (5)

  • Figure 1: (Color online) Finding the erasure pseudothreshold for square lattice toric codes. Symbols show the homological error probability (\ref{['eq:prob-E']}) evaluated numerically for different graphs labeled by the number of edges $n=2d^2$, with distances $d$ ranging from $60$ to $220$, plotted as a function of open edge probability $p$. As expected, beautiful crossing point very close to $p_E^0=1/2$ is observed. Lines are the polynomials $f_n(\xi)=A_0+A_{1n}\xi+A_{2n}\xi^2+\ldots$ of $\xi=p-p^0$ of degree 6 obtained by fitting the data collectively in the range $0.49\le p\le 0.51$. The vertical dashed line indicates the square lattice percolation threshold $p_{\rm c}=1/2$.
  • Figure 2: (Color online) Top: as in Fig. \ref{['fig:PE-4-4']} but for the hyperbolic code family $\{5,5\}$. The green arrow indicates the position of the crossing point found by the fit; it is significantly below the percolation threshold for the corresponding infinite lattice (vertical red dashed line). In addition, the data for the graph with $n=15\,350$ is shifted upward, which we associate with a slightly smaller ratio $d/\ln n$, see Fig. \ref{['fig:dist']}. This is verified in the bottom plot, where an additional vertical shift proportional to $\ln n/d$ is added, which substantially improves the convergence at the crossing point.
  • Figure 3: Homological distance $d\equiv d_Z$ associated with non-trivial cycles for optimal graphs in $\{7,3\}$ and $\{5,5\}$ families vs. the graph size $n$ (number of edges) with the logarithmic scale. Numbers also indicate the graph sizes. Smaller relative distances $d/\ln n$ result in larger erasure probabilities in Figs. \ref{['fig:pE-hyp55']} and \ref{['fig:pE-hyp73']} (top); this can be compensated to some extend by using the correction term as in bottom plots in Figs. \ref{['fig:pE-hyp55']} and \ref{['fig:pE-hyp73']}.
  • Figure 4: (Color online) As in Fig. \ref{['fig:pE-hyp55']} but for the hyperbolic code family $\{7,3\}$. The convergence at the crossing point is much better than that in Fig. \ref{['fig:pE-hyp55']} (top), which we associate with substantially higher ratios $d/\ln n$ for the graphs in this family. Bottom plot: addition of the additional vertical shift $B\ln n/d$ causes a substantial shift of the crossing point position without visibly improving the convergence.
  • Figure 5: (Color online) Using the random-graph-like scaling for locating the percolation transition for hyperbolic graphs in the $\{5,5\}$ (top) and $\{7,3\}$ (middle) families, and for degree-$5$ random graphs (bottom). Values of the open bond probability $p$ where the expected size of the largest cluster equals $\omega n^{2/3}$ are plotted as a function of $n^{-1/3}$, for values of $\omega$ as indicated. Here $n$ is the number of edges in the graph. The lines intersect close to the percolation transition point, as indicated by horizontal dashed lines.

Theorems & Definitions (34)

  • Lemma \oldthetheorem
  • Lemma \oldthetheorem
  • Lemma \oldthetheorem
  • Theorem \oldthetheorem
  • Theorem \oldthetheorem
  • Theorem \oldthetheorem
  • Corollary \oldthetheorem
  • Example \oldthetheorem: Anisotropic square-lattice toric codes
  • Theorem \oldthetheorem
  • Theorem \oldthetheorem
  • ...and 24 more