1: \begin{abstract}

2: Symmetric nonnegative matrix factorization (SymNMF) has important

3: applications in data analytics problems such as document clustering, community

4: detection and image segmentation. In this paper, we propose a novel nonconvex

5: variable splitting method for solving SymNMF. The proposed algorithm is

6: guaranteed to converge to the set of Karush-Kuhn-Tucker (KKT) points of the

7: nonconvex SymNMF problem. Furthermore, it achieves a global sublinear

8: convergence rate. We also show that the algorithm can be efficiently

9: implemented in parallel. Further, sufficient conditions are provided which

10: guarantee the global and local optimality of the obtained solutions. Extensive

11: numerical results performed on both synthetic and real data sets suggest that

12: the proposed algorithm converges quickly to a local minimum solution.

13: 

14: \begin{keywords}

15: Symmetric nonnegative matrix factorization, Karush-Kuhn-Tucker points,

16: variable splitting, global and local optimality, clustering

17: \end{keywords}

18: \end{abstract}