On the Jacobian conjecture