[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Levenberg Marquadt parameter limits

In article <7dbmvv$tes$1@nnrp1.dejanews.com>,
  mws@jet.uk wrote:
> Hi, I'm using a L-M Chi-squared minimisation routine to estimate some
> parameters in my experiment.

Thanks to all the people who replied to my posting about this question. I was
trying to avoid re-parameterising the equations as this is a nightmare with
derivatives, but it was to be my next step if nothing better came along.
Upper and lower limits would be trickier still. FYI, I've now found a great
book by Gill & Wharton & alii (Practical Optimisation??) 1981 which suggests
that LM fit is NOT for constrained parameters, and I should try another
method (un-named in the book). This method in this method defines constraints
which define the 'feasible' parameter space.

I've found that the E04 NAG routines do this work, so I'm going to give it a
go with one of them. I was sent a link to a neat NIST USA site:
http://math.nist.gov/gams/ which is very good at helping one decide which
optimisation routine to use. This pointed me to the same routine I had chosen
(yesterday) the hard way.


-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own