Electromagnetic-transient (EMT) as well as Dynamic-Phasor (DP) simulations of power system stability events need repeated factorization of large sparse Modified Nodal Analysis (MNA) matrices whenever topology changes occur—such as short circuit faults, line protection tripping and voltage induced tripping of generators or loads.

Traditional approaches rebuild the full system matrix at each event or precompute system matrices based on anticipated switching events.

In this paper we propose a network tearing (diakoptics) approach that partitions the system into subnetworks, recomputing only the MNA blocks associated with topology changes. Our method extends classical tearing (Dommel, 1979) and Multi Area Thévenin Equivalent (MATE) techniques (Huang & Vittal, 2016; Huang et al., 2017). Previous studies have already shown that parallelizing simulation steps within the diakoptics framework can significantly improve performance. In this work, we show that diakoptics offers performance benefits even without parallelization, as only the matrix blocks affected need to be recomputed for topology changes. We validate our approach against full-system MNA solutions for both EMT and DP simulations and present results quantifying performance improvements and scalability across different network sizes and partitioning granularities.