An acceleration of gradient descent algorithm with backtracking for unconstrained opti\-mi\-za\-tion (Q2502232)

The author considers the unconstrained optimization problem where the objective function is convex and continuously differentiable. He focuses on the gradient descent algorithm with backtracking, and describes an improved version which accelerates convergence. In particular, the acceleration is based on a modification of the steplength by multiplying it with a positive parameter. The resulting algorithm is proven to be linearly convergent, and numerical experimentation in section 4 shows a significant improvement in performance.