1: \begin{abstract}

2:     This paper studies the multi-agent average consensus problem under

3:     the requirement of differential privacy of the agents' initial

4:     states against an adversary that has access to all the messages.

5:     We first establish that a differentially private consensus

6:     algorithm cannot guarantee convergence of the agents' states to

7:     the exact average in distribution, which in turn implies the same

8:     impossibility for other stronger notions of convergence.  This

9:     result motivates our design of a novel differentially private

10:     Laplacian consensus algorithm in which agents linearly perturb

11:     their state-transition and message-generating functions with

12:     exponentially decaying Laplace noise.  We prove that our algorithm

13:     converges almost surely to an unbiased estimate of the average of

14:     agents' initial states, compute the exponential mean-square rate

15:     of convergence, and formally characterize its differential privacy

16:     properties. We show that the optimal choice of our design

17:     parameters (with respect to the variance of the convergence point

18:     around the exact average) corresponds to a one-shot perturbation

19:     of initial states and compare our design with various counterparts

20:     from the literature.  Simulations illustrate our results.

21:   \end{abstract}

22: