[Ifeffit] Re: differential evolution algorithm (fwd)

[Grant Bunker]

Carlo suggested that I submit this (edited) to the ifeffit list... gb
---------- Forwarded message ----------
Date: Thu, 2 Dec 2004 21:40:54 -0600 (CST)
From: Grant Bunker <bunker at agni.phys.iit.edu>
To: Carlo U. Segre <segre at iit.edu>
Cc: Bruce Ravel <ravel at phys.washington.edu>,
     Grant Bunker <bunker at biocat1.iit.edu>
Subject: Re: differential evolution algorithm

Hi, Carlo - DE is a very nice and simple genetic algorithm for finding
global minima. As far as I know I have been the only person applying it to
XAFS. It might be overkill for the application though - is this just for
finding maximum in the edge derivative?  If it is, it's probably simpler
just to smooth the data (e.g. with binomial smoothing which you can do by
repeated convolution Y(n)=y(n+1)+2*y(n)+y(n-1))/4), followed by a
derivative. I've done it that way before (or use smoothing splines) in the
distant past and it worked fine. In the edge-finding hack in the mathematica
notebook (for automated data reduction) I sent you I just used differential
evolution out of laziness, but it's very good for more difficult fitting
problems like the stuff we have been doing with the ab initio DWFs in enzymes.

It's arguable whether using the max derivative is the best way to
choose E0 anyway, but that's a different story.

thanks - grant

ps I googled "differential evolution" and uncovered a good web site at
berkeley. Probably it has references to the original article by storn and
price. A couple of years ago I implemented it in Mathematica code for
global fitting calling feff as an external "function". The intrinsic
NMinimize function in Mathematica also has it available as a minimization
method in purely numerical (nonsymbolic) calculations.

http://www.icsi.berkeley.edu/~storn/code.html

On Thu, 2 Dec 2004, Carlo U. Segre wrote:

>
> I am somewhat ignorant.  Grant uses the builtin function from Mathematica.
> Perhaps he can give a bit more information.  Grant?
>
> Carlo
>
> On Thu, 2 Dec 2004, Bruce Ravel wrote:
>
> >
> >> Grant has been using a differential evolution algorithm to take case of
> >> noisy data.  Perhaps this could help?
> >
> > Carlo,
> >
> > Is there a reference for this in the exafs conmtext?  I'd like to read
> > up on it.
> >
> > B
> >
> >
> >
>
> --
> Carlo U. Segre -- Professor of Physics
> Associate Dean for Special Projects, Graduate College
> Illinois Institute of Technology
> Voice: 312.567.3498            Fax: 312.567.3494
> Carlo.Segre at iit.edu    http://www.iit.edu/~segre
>