1: \begin{abstract}

2:   We propose a weight design method to increase the convergence rate of

3:   distributed consensus. Prior work has focused on symmetric weight

4:   design due to computational tractability. We show that with proper

5:   choice of \emph{asymmetric} weights, the convergence rate can be

6:   improved significantly over even the symmetric optimal design. In

7:   particular, we prove that the convergence rate in

8:   a lattice graph can be made independent of the size of the graph

9:   with asymmetric weights. We then use a

10:   Sturm-Liouville operator to approximate the graph Laplacian of more

11:   general graphs. A general weight design method is proposed based on

12:   this continuum approximation. Numerical computations show that the

13:   resulting convergence rate with asymmetric weight design is improved

14:   considerably over that with symmetric optimal weights and

15:   Metropolis-Hastings weights.

16: \end{abstract}

17: