Reconstruction/Non-reconstruction Thresholds for Colourings of General Galton-Watson Trees

TL;DR

This paper investigates the reconstruction thresholds for coloring models on Galton-Watson trees with general offspring distributions, extending known results from regular trees to more complex, probabilistic tree structures.

Contribution

It generalizes the reconstruction threshold analysis from regular trees to Galton-Watson trees with arbitrary offspring distributions, linking thresholds to offspring distribution concentration properties.

Findings

Reconstruction thresholds are derived for a wide family of offspring distributions.

Threshold for distributions with expected offspring d is d/ln(d) under weaker concentration conditions.

Results imply thresholds for coloring on G(n,d/n) graphs based on Galton-Watson tree analysis.

Abstract

The broadcasting models on trees arise in many contexts such as discrete mathematics, biology statistical physics and cs. In this work, we consider the colouring model. A basic question here is whether the root's assignment affects the distribution of the colourings at the vertices at distance h from the root. This is the so-called "reconstruction problem". For a d-ary tree it is well known that d/ln (d) is the reconstruction threshold. That is, for k=(1+eps)d/ln(d) we have non-reconstruction while for k=(1-eps)d/ln(d) we have. Here, we consider the largely unstudied case where the underlying tree is chosen according to a predefined distribution. In particular, our focus is on the well-known Galton-Watson trees. This model arises naturally in many contexts, e.g. the theory of spin-glasses and its applications on random Constraint Satisfaction Problems (rCSP). The aforementioned study focuses on Galton-Watson trees with offspring distribution B(n,d/n), i.e. the binomial with parameters n and d/n, where d is fixed. Here we consider a broader version of the problem, as we assume general offspring distribution, which includes B(n,d/n) as a special case. Our approach relates the corresponding bounds for (non)reconstruction to certain concentration properties of the offspring distribution. This allows to derive reconstruction thresholds for a very wide family of offspring distributions, which includes B(n,d/n). A very interesting corollary is that for distributions with expected offspring d, we get reconstruction threshold d/ln(d) under weaker concentration conditions than what we have in B(n,d/n). Furthermore, our reconstruction threshold for the random colorings of Galton-Watson with offspring B(n,d/n), implies the reconstruction threshold for the random colourings of G(n,d/n).