[R] How to define stopping criterium for Optim with L-BFGS-B

[Prof Brian Ripley]

On Wed, 23 Jun 2004, ZABALZA-MEZGHANI Isabelle wrote:

> I am using optim with a L-BFGS-B method to minimize a function. As I've
> understood, the way to specify a tolerance for stopping optimization is
> through "factr" argument.
> My function, is by construction, minimal when equal to 1. I wonder if there
> is any way to pass this info to "optim". 

No, as the problem is assumed to be scaled so this is approximately true.

> If not, how "factr" argument works
> (I am quite confused about the relationship between this argument and the
> Machine eps). Please, could someone give me an example ?

The (absolute) convergence tolerance is factr times the Machine eps, and
the help page does already give you an example.  You can read the
references for further details.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595