Curve Fitting, why is the increment of the points so large (aka resolution of points so low)

Share icon

Tue, 09/20/2011 - 07:56 am

Hey,

I'm trying to fit a tri-exponential over a wide range of data. I've noticed that the resolution of the fitting curve depends on how large a range is fitted over... Is there anyway to improve the resolution of the fitting curve so that it fits the data better? Or does it even matter?

Thanks!!
It doesn't really matter- it's just a wave full of fitted Y values made by executing the fit function with the fitted coefficients. It is always possible to execute the function at arbitrary X values.

However....

In the Curve Fit dialog, in the Output Options tab, you can set the number of points used in the fit curve.

It is likely that your tri-exponential fit really needs a high density of points at small X and low density at large X. If you make the X axis a log axis before fitting, Igor will make the fit curve an XY pair where the X values have exponential spacing. That will give a smooth curve on a log axis, and will have higher density at small X. This may not work well if your data don't start near the origin.

You can always make your own fit curve with any X values you want:
Make/N=<your choice> fitcurveX, fitcurveY
fitcurveX = <some expression that fills the wave with values that you like>

Now copy the example assignment out of the curve fit report printed in the history, paste it into the command line and edit it to use the waves you just created. For instance, if you had fit a double exponential using the built-in fit function, you might find something like this in the history report:
  fit_junk= W_coef[0]+W_coef[1]*exp(-W_coef[2]*x)+W_coef[3]*exp(-W_coef[4]*x)

Replace all instances of "x" with "fitcurveX" and replace "fit_junk" with "fitcurveY". Naturally, your names will be different than "fit_junk", "fitcurveX" and "fitcurveY".

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com

September 20, 2011 at 08:59 am - Permalink