1: \begin{abstract}

2: Personalized PageRank Vectors are widely used as fundamental graph-learning tools for learning graph embeddings, training graph neural networks, and detecting anomalous spammers. The well-known \textit{local} \textsc{FwdPush} algorithm \cite{andersen2006local} approximates PPVs and has a sublinear rate of $\mc{O}\big(\frac{1}{\alpha\eps}\big)$. A recent study \cite{wu2021unifying} found that when high precision is required, \textsc{FwdPush} is similar to the power iteration method, and its run time is pessimistically bounded by $\mc{O}\big(\frac{m}{\alpha} \log\frac{1}{\eps}\big)$. This paper looks closely at calculating PPVs for both directed and undirected graphs. By leveraging the \textit{linear invariant property}, we show that \textsc{FwdPush} is a variant of Gauss-Seidel and propose a Successive Over-Relaxation based method, \textsc{FwdPushSOR} to speed it up by slightly modifying \textsc{FwdPush}. Additionally, we prove \textsc{FwdPush} has \textit{local linear convergence} rate $\mc{O}\big(\tfrac{\vol(\mc S)}{\alpha} \log\tfrac{1}{\eps}\big)$ enjoying advantages of two existing bounds. 

3: We also design a new local heuristic push method that reduces the number of operations by 10-50 percent compared to \textsc{FwdPush}. For undirected graphs, we propose two momentum-based acceleration methods that can be expressed as one-line updates and speed up nonacceleration methods by $\mathcal{O}(1/\sqrt{\alpha})$. Our experiments on six real-world graph datasets confirm the efficiency of \textsc{FwdPushSOR} and the acceleration methods for directed and undirected graphs, respectively.

4: \end{abstract}

5: