Re: gnumeric-list Digest, Vol 59, Issue 21

["Andreas J. Guelzow" <aguelzow pyrshep ca>]

On Fri, 2009-03-27 at 08:54 -0600, Jim Martin wrote:
> gnumeric-list-request gnome org wrote: 
> > Date: Thu, 26 Mar 2009 11:12:55 -0600
> > From: "Andreas J. Guelzow" <aguelzow math concordia ab ca>
> > Subject: Re: gnumeric-list Digest, Vol 59, Issue 21
> >
> > According to the Gnumeric solver dialog, the solver uses a simplex
> > algorithm to optimize the function. I really don't think that is the
> > right tool for your kind of problem.
> >
> > Which kind of algorithm is used by OpenOffice Calc and/or Excel? 
> >   
> According to the microsoft web site, Solver in Excel uses the
> "Generalized Reduced Gradient Algorithm". The solver plugin is written
> by Frontline Systems (www.solver.com)
>
> The Solver for Nonlinear Programming extension for OpenOffice consists
> of two independent algorithms: Differential Evolution and Particle
> Swarm Optimization. For more information see:
> http://wiki.services.openoffice.org/wiki/NLPSolver

Great thanks,

> > (I doubt that there is a general solver that will "converge in seconds" for an arbitrary problem.)
> Well Andreas, I certainly respect your academic skepticism. Indeed,
> there is no reason to either accept or doubt my claim. 

I really did not doubt the claim that for your problem you saw
convergence in seconds. I was commenting about "an arbitrary problem".
The fact that your problem converged in seconds simply tells me that the
algorithm that was used happened to be well suited for your problem, it
doesn't tell me anything about the quality of the algorithm in general,
or it s speed with repsect to arbitrary problems. That's why I was
asking whether you knew which algorithms were used. 

> Rather, I encourage you to check the veracity of my report yourself
> (presuming you have access to a computer with excel).

Since I do not doubt your report (with respect to your specific problem
or type of problem) and do not have any easy access to a computer with
Excel (since that would presuppose easy access to a machine with MS
Windows or MacOS) I will just take your word (which I never doubted).

>   The article and the spreadsheet are available on line at
> http://sportsci.org/2009/sjejcm.htm  Once you go the article, you will
> see a link to the spreadsheet. So please feel free (indeed obligated
> by your publicly stated doubt of my report) to download that
> spreadsheet, set the Fourier coefficients to some new values, and then
> run solver yourself. Please let us know how quickly it converges for
> you. When I reset the coefficients for the 1st order approximation to
> zero solver finds the solution almost instantaneously. Perhaps that is
> somehow related to the fact that solver has done the approximation
> before on my computer and has access to that solution history. So
> please try it from scratch on your computer. 

I'll see whether I can track down access to an MS Windows machine with
Excel and let you know.

Andreas
>

-- 
Andreas J. Guelzow, PhD, FTICA
Coordinator, Mathematical & Computing Sciences
Concordia University College of Alberta