Scaling in fmincon()

Hi all,

I have been using the Matlab constrained optimization fucntion(). Lately, I
have been observing that the speed and the success of the optimization
depends on the "scale" of the problem. For example, I have observed that
when I divide the objective function by a constant so as to make the
function values smaller, the optimization converges significantly faster.

I see in the literature that scaling is an important factor in optimization,
and optimization softwares normally have automatic procedures for
transforming the user-specified problem to a well-scaled one.

So, does anyone know if Matlab fmincon() does any scaling at all? Has anyone
experienced scaling problems before with fmincon()?

Thanks
Fijoy


.

Relevant Pages

  • Re: fmincon not updating
    ... One scaling approach is to map the original optimization variable space to one where the lower bound is 0 and the upper bound is 1 for each variable. ... In some cases a nonlinear transformation is required so that the sensitivity with respect to optimization variables is fairly uniform over the optimization domain; the optimization algorithm gets stuck if it's not. ... In addition to scaling the variables, it's important that the objective and constraint functions are scaled appropriately as well. ...
    (comp.soft-sys.matlab)
  • Re: Optimization
    ... >Peter Spellucci wrote: ... Scaling nonlinear programs. ... Nonlinear scaling algorithms for unconstrained optimization. ...
    (sci.math.num-analysis)
  • Re: fmincon optimization issue
    ... Scaling up the variables does give me some better result. ... For some other reason, I need to adjust some other parameter of the optimization, then I keep getting result that seems pretty bad, which I assume is due to bad starting point, and I am trying to use John's rmsearch.m script. ...
    (comp.soft-sys.matlab)