Solving quadratic programming problems with linear Hopfield networks

We consider a linear Hopfield network for solving quadratic programming problems with equation constraints. The problem is reduced to the solution of ordinary linear differential equations with arbitrary square matrix. Because of some properties of this matrix special methods are required for good convergence of the system. After some comparative studies of neural network models for solving this problem we suggest a modified system with doubling of the number of variables. The new system is very simple in implementation on the linear Hopfield network and demonstrates sufficiently good convergence to the solution