1: \begin{abstract}

2:       In this work, we consider the decentralized optimization problem in which a network of $n$ agents, each possessing a smooth and convex objective function, wish to collaboratively minimize the average of all the objective functions through peer-to-peer communication in a directed graph.

3:     To solve the problem, we propose two accelerated Push-DIGing methods termed $\NAPD$ and $\NAPDSC$ for minimizing non-strongly convex objective functions and strongly convex ones, respectively.

4:     We show that $\NAPD$ and $\NAPDSC$ respectively converge at the rates $O\pr{\frac{1}{k^2}}$ and $O\pr{\pr{1 - C\sqrt{\frac{\mu}{L}}}^k}$ up to constant factors depending only on the mixing matrix.

5:     To the best of our knowledge, $\NAPD$ and $\NAPDSC$ are the first decentralized methods to achieve provable acceleration over unbalanced directed graphs.

6:     Numerical experiments demonstrate the effectiveness of both methods.

7: 

8: %    In this work, we consider the decentralized optimization problem in which each agent owns a smooth convex objective function and can communicate directly only with its neighbors in a directed graph.

9: %   Since the construction of doubly stochastic matrices given the directed communication topology is usually costly and even impractical,    we propose the accelerated Push-DIGing methods $\NAPD$ and $\NAPDSC$ for non-strongly convex functions and strongly convex functions, which use only column stochastic matrices.    $\NAPD$ and $\NAPDSC$ are the first decentralized methods over unbalanced graphs to achieve convergence rates of $O\pr{\frac{1}{k^2}}$ and $O\pr{\br{1 - C\sqrt{\frac{\mu}{L}}}^k}$ for non-strongly convex and strongly convex functions, respectively.

10:  %   Numerical experiments demonstrate the effectiveness of both methods.

11: 

12: 

13: \end{abstract}