COMPUTING A CLOSEST BIFURCATION INSTABILITY IN MULTIDIMENSIONAL PARAMETER SPACE

Engineering and physical systems are often modeled as nonlinear differential equations with a vector lambda of parameters and operated at a stable equilibrium. However, as the parameters lambda vary from some nominal value lambda0, the stability of the equilibrium can be lost in a saddle-node or Hopf bifurcation. The spatial relation in parameter space of lambda0 to the critical set of parameters at which the stable equilibrium bifurcates determines the robustness of the system stability to parameter variations and is important in applications. We propose computing a parameter vector lambda* at which the stable equilibrium bifurcates which is locally closest in parameter space to the nominal parameters lambda0. Iterative and direct methods for computing these locally closest bifurcations are described. The methods are extensions of standard, one-parameter methods of computing bifurcations and are based on formulas for the normal vector to hypersurfaces of the bifurcation set. Conditions on the hypersurface curvature are given to ensure the local convergence of the iterative method and the regularity of solutions of the direct method. Formulas are derived for the curvature of the saddle node bifurcation set. The methods are extended to transcritical and pitchfork bifurcations and parametrized maps, and the sensitivity to lambda0 of the distance to a closest bifurcation is derived. The application of the methods is illustrated by computing the proximity to the closest voltage collapse instability of a simple electric power system.