No label defined (Q4515930)

The author investigates the minimization problem \[ f(x)\to \inf,\quad x\in Q\subset E^n,\tag{1} \] where \(Q\) is a convex closed set, \(f(x)\) is a smooth convex function whose gradients satisfy the Lipschitz condition. It is proposed to solve this problem by two-step projection method. In order that to obtain the next approximation at the minimum point \(x^*\) the gradient of function \(f(x)\) is used in some intermediate point \(z^k\), rather than at point~\(x^k\) \[ x^{k+1} = P_Q[x^k+\alpha_k y^k - \gamma_kf'(z^k)],\quad k=0,1,2,\dots, \] where \( x^{-1}=x^0\), \( \alpha_k\) and \(\gamma_k\) are the method parameters, \( y^k=x^k-x^{k-1}\), \( z^k=x^k+\alpha_ky^k\). The results of numerical experiments are presented.