On the solution of systems of quadratic equations

The paper addresses the classical problem of solving a system of quadratic algebraic equations. To find a solution, we apply a variational approach of reducing the original problem to an optimization problem with cost functions represented by the difference of convex functions (DC functions). In this case, the optimization problem turns out to be nonconvex and nonsmooth. Using the known properties of DC functions, we reduce the main optimization problem to a smooth DC minimization problem with inequality constraints. To solve the latter problem, first, a special local search method (SLSM) is applied. The convergence of the method is proved and stopping criteria are established. The testing of the SLSM on systems of equations with quadratic data shows its sufficient efficiency, even when compared with known special applied software packages. Global search procedures are built on the basis of global optimality conditions, which allow us to “jump out” of local solutions and stationary and critical points. The global search method is tested. The comparative efficiency of the developed approach is proved by the successful solution of all test systems of quadratic equations for problems of large dimension (with the number of equations and variables up to 1000). At the same time, the standard applied software packages fail to solve large-dimensional problems for the most part of test examples.