Towards Geometric Completion of Differential Systems by Points

Numerical Algebraic Geometry represents the irreducible components of algebraic varieties over C by certain points on their components. Such witness points are efficiently approximated by Numerical Homotopy Continuation methods, as the intersection of random linear varieties with the components. We outline challenges and progress for extending such ideas to systems of differential polynomials, where prolongation (differentiation) of the equations is required to yield existence criteria for their formal (power series) solutions. For numerical stability we marry Numerical Geometric Methods with the Geometric Prolongation Methods of Cartan and Kuranishi from the classical (jet) geometry of differential equations. Several new ideas are described in this article, yielding witness point versions of fundamental operations in Jet geometry which depend on embedding Jet Space (the arena of traditional differential algebra) into a larger space (that includes as a subset its tangent bundle). The first new idea is to replace differentiation (prolongation) of equations by geometric lifting of witness Jet points. In this process, witness Jet points and the tangent spaces of a jet variety at these points, which characterize prolongations, are computed by the tools of Numerical Algebraic Geometry and Numerical Linear Algebra. Unlike other approaches our geometric lifting technique can characterize projections without constructing an explicit algebraic equational representation. We first embed a given system in a larger space. Then using a construction of Bates et al., appropriate random linear slices cut out points, characterizing singular solutions of the differential system.