1: \begin{abstract}

2: We analyze the convergence of decentralized consensus

3: algorithm with delayed gradient information across the network. 

4: The nodes in the network privately hold parts of the objective 

5: function and collaboratively solve for the consensus

6: optimal solution

7: of the total objective while they can only communicate with their immediate neighbors.

8: In real-world networks, it is often difficult and sometimes impossible to 

9: synchronize the nodes, and therefore they have to use stale gradient information 

10: during computations.

11: We show that, as long as the random delays are bounded in expectation

12: and a proper diminishing step size policy is employed,

13: the iterates generated by decentralized gradient descent method 

14: converge to a consensual optimal solution.

15: Convergence rates of both objective and consensus are derived.

16: Numerical results on a number of synthetic problems and real-world seismic tomography datasets

17: in decentralized sensor networks are presented to show the performance of the method.

18: \bigskip

19: 

20: \noindent

21: \textbf{Key words.} Decentralized consensus, delayed gradient, stochastic gradient, decentralized networks.

22: \bigskip

23: 

24: \noindent

25: \textbf{AMS  subject classifications.} 65K05, 90C25, 65Y05.

26: \end{abstract}

27: