Rigidity of 3-colorings of the discrete torus (Q1650127)

The authors study $3$-colourings of the $d$-dimensional discrete torus $ \mathbb{T}_n^d := \left(\mathbb{Z} / n \mathbb{Z} \right)^d$, where $n \in 2\mathbb{N}$, so that $\mathbb{T}_n^d$ is bipartite. The main result is captured in Theorem 1.1. Before stating that result, recall that a proper $3$-colouring of $\mathbb{T}_n^d$ is any function $f: \mathbb{T}_n^d \rightarrow \{0, 1, 2\}$ such that, if $v$ and $w$ are adjacent in $\mathbb{T}_n^d$, then $f(v) \neq f(w)$. \par Theorem 1.1 states that, if $d$ is a big number (high-dimensional case), then a uniformly chosen proper $3$-colouring of $\mathbb{T}_n^d$ is nearly constant on one of the two bipartition classes of $\mathbb{T}_n^d$. It can be interpreted as follows. In high dimensions, for each colour, there is a partite class on which this colour hardly appears, or equivalently, one of the partite classes is dominated by only one colour. \par To prove Theorem 1.1, the authors establish a connection between proper $3$-colourings and homomorphism height functions (HHF, functions on the vertices of bipartite graph taking integer values and constrained to have adjacent vertices take adjacent integer values), so they can apply the results regarding height functions from the second author [Ann. Probab. 45, No. 3, 1351--1447 (2017; Zbl 1377.82021)]. Before these results (on height functions) could be applied, the authors have to deal with one technical issue, that is, that the connection they establish is not bijective. To overcome this obstacle, they provide Theorem 1.2, which states that this connection is `almost' bijective when the dimension $d$ is sufficiently large. \par The authors also present a statistical physics interpretation of their work, namely, that the proper $3$-colouring model coincides with the zero-temperature case of the anti-ferromagnetic $3$-state Potts model. Then they state the equivalent formulation in terms of $3$-colourings of the positive-temperature version of the Potts model and leave the analogue of Theorem 1.1 in this setting as an open problem.