1: \begin{abstract}

2: We study local Markov chains for sampling 3-colorings of the discrete torus

3: $T_{L,d}=\{0,\ldots, L-1\}^d$.

4: We show that there is a constant $\rho \approx .22$ such that for

5: all even $L \geq 4$ and $d$ sufficiently large, certain local Markov chains require

6: exponential time to converge to equilibrium.  More precisely,

7: if $\cM$ is a Markov chain on the set of proper $3$-colorings

8: of $T_{L,d}$ that updates the color of at most $\rho  L^d$ vertices

9: at each step and whose stationary distribution is uniform, then the

10: convergence to stationarity of $\cM$ is exponential in

11: $L^{d-1}$.

12: Our proof is based on a conductance argument that builds on sensitive

13: new combinatorial enumeration techniques.

14: \end{abstract}