estimating parameters in L-M


andbeck
04-11-2005, 03:54 AM
Hi

I have managed to build a fully executable program that uses the Marquardt method.

This probably sounds like a stupid question but once I have my data to be fitted for and plot it on a graph, how do I estimate what the initial values for the parameters should be just by looking at the graph, before I carry out the fitting? The non-linear function I am trying to fit for is:

y = a[0]*exp(-x/a[1]) + a[2]*exp(-x/a[3]) + a[4]*exp(-x/a[5])

i.e. it is the sum of three decaying exponentials.

When I plot my experimental data I obtain a nice exponential decay curve which should hopefully fit to the above form, so how do I estimate what a[0]...a[5] should be before I put them into the program from the data on my graph?

Many thanks in advance,

Beck