The system was energy-minimized 6 times switching alternatively between runs using steepest descent gradients or Polak-Ribière conjugate gradients until convergence to machine precision [16].