Date: Jul 18, 2013 5:11 PM
Author: Sargondjani
Subject: Re: RelLineSrchBnd option for fmincon

Thanks Alan. The thing is, I read it carefully, but I guess I didn't understand it because I didn't know what is meant with 'displacement' and 'line search step length'. 

After reading this documentation
http://www.mathworks.com/help/optim/ug/constrained-nonlinear-optimization-algorithms.html#f26965

i found the following:

"Line Search and Merit Function: The solution to the QP subproblem produces a vector dk, which is used to form a new iterate" and then "The step length parameter ?k is determined in order to produce a sufficient decrease in a merit function."

But I don't really understand.

In fact I did find out that RelLineSrchBnd does affect the stepsize in the iteration, but not as much as I expected. For example, when I set RelLineSrchBnd=0.01 then the change in the x-values in the first iteration only seem to get smaller by factor 10 (where I expected a factor 100).

Maybe you can explain in more simple words what RelLineSrchBnd does? What it means in practice? And why is the stepsize in the iteration only affected so little?)

Many thanks in advance!