Computational Science Stack Exchange is a question and answer site for scientists using computers to solve scientific problems. It only takes a minute to sign up.
Sign up to join this community
Can anyone point me in the direction of literature pertaining to the selection of the coefficient c2 in the curvature condition in the wolfe conditions when performing a line search?
In Nocedal, he says that c2 = 0.9 for quasi Newton methods, and that c2 = 0.1 for non linear conjugate gradient methods.
I have not been able to find any literature pertaining to this.
Thank you!
There is no magic to these parameters. Line search is a tool to get from areas where a full-step Newton's method doesn't converge into those areas where it does converge. For this, it is not necessary to be absolutely optimal in your choice of step length acceptance criteria, the parameters in these criteria, or the exactness of the Newton direction. They're all just tools to get to those places where you can take a full step.
What I mean by all this is that these parameters are somewhat heuristic choices. The values Nocedal and Wright give are widely used, but there is no fundamental theory to back these choices up.
