This paper proposes a sensitivity matrix-based formulation methodology for decentralized optimal reactive power dispatch problems on multi-region interconnected power systems. A large power system is divided into multi-region, with region operators agreeing on the progression of variables and determining optimal local control variables using the local system model and a communication network with its adjacent control regions. The constrained objective function is substituted with a sequence of unconstrained sub-problems using the augmented Lagrangian approach to achieve global optimization. To specify control variables, such as reactive power injections, and transformer tap positions, a linearized objective function with a set of local constraints must be addressed in each region. A nonlinear optimization algorithm using function fmincon in Matlab is utilized to solve the mathematical model effectively and is applied to the modified IEEE 30-bus system to demonstrate the validity and effectiveness of the proposed method.