The computational bottleneck for large nonlinear AC power flow problems using Newtonâ€™s method is the solution of the linear system at each iteration. We present a parallel linear solution scheme using the Krylov subspace-based iterative solver GMRES preconditioned with overlapping restricted additive Schwarz method (RASM) that shows promising speedup for this linear system solution. This paper evaluates the performance of RASM with different amounts of overlap and presents its scalability and convergence behavior for three large power flow problems consisting of 22,996, 51,741, and 91,984 buses respectively.

%B The International Conference for High Performance Computing, Network, Storage and Analysis (SC 2013) %C Denver, CO %G eng %1 http://www.mcs.anl.gov/papers/P5021-0913_1.pdf