Two modified Dai-Yuan nonlinear conjugate gradient methods (Q1014353)

The author improves the iterative method \[ x_{k+1}=x_k+\alpha_k d_k;\;d_0=-g_k,\;d_k=-g_k+\beta_k d_{k-1},\;k>0;\;g(x)=f^{\prime}(x),\;g_k=g(x_k) \] for solving smooth unconstrained optimization problems \(\min_{x \in\mathbb R^n} f(x)\). In the original version of the method, \[ \beta_k=\| g_k \|^2/d^T_{k-1}y_{k-1},\;y_{k-1}=g_k-g_{k-1}, \] and the stepsize \(\alpha_k>0\) satisfies \[ f(x_k+\alpha_k d_k) \leq f(x_k)+\delta \alpha_k g_k^T d_k,\;d^T_k g(x_k+\alpha_k d_k) \geq \sigma d^T_k g_k, \] where \(0<\delta<\sigma<1\). The modified method converges globally for nonconvex functions such that the level set \(\Omega=\{ x\in \mathbb R^n: f(x) \leq f(x_0) \}\) is bounded and \(\| g(x)-g(y)\| \leq L\| x-y \|\) for all \(x,y\) from a neighborhood of \(\Omega\).