1: \begin{abstract} 

2: We consider distributed optimization over a graph which aims to optimize a global objective formed by a sum of coupled local convex functions via local computation and communication. We propose Bregman parallel direction method of multipliers (PDMM) based on a generalized consensus protocol, named mirror Markov mixing. We establish the global convergence and \(O(1/T)\) iteration complexity of Bregman PDMM, along with its \(O(n/\ln n)\) improvement over an existing PDMM where \(n\) is the dimensionality of solution variable. In addition, we show that its performance can be enhanced by optimizing the spectral gap of Markov mixing matrix. We also demonstrate its effectiveness via a numerical example. 

3: \end{abstract}