Example of turning from ascent to descent in a fixed stepsize gradient method

29 Views Asked by At

We have limited understanding of the objective function since it is implicit from the input variables. When applying the classical gradient ascent method in the fixed stepsize, the numerical shows that the objective function was close to the maximum in few iterations, but keep decreasing after that. Is there any explicit example to demonstrate such a behavior? I can understand it diverges from the maximum due to large stepsize, but will expect the objective function increases after decreasing for a while.